You are on page 1of 88

GENERAL RELATIVITY.

MATH3443

KOMISSAROV S.S

2009
2
Contents

Contents 2

1 Introduction 5

2 From Euclidean space to surfaces and metric manifolds 11


2.1 Metric form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.1 The notion of metric form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.2 Metric forms of surfaces: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.3 Lengths of curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.4 Coordinate transformations: . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Vectors, bases, and components of vectors . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.1 Coordinate bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.2 Coordinate transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Metric form and the scalar product . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Geodesics and the variational principle . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4.1 Euler-Lagrange Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4.2 Geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4.3 Examples of geodesics: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.5 Non-Euclidean geometry of a Euclidean sphere . . . . . . . . . . . . . . . . . . . . . 20
2.6 Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.7 Vectors as operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.7.1 Basic idea . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.7.2 Coordinate transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.7.3 Magnitudes of vectors and the scalar product . . . . . . . . . . . . . . . . . . 23

3 Tensors 25
3.1 Tensors as operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.1.1 1-forms as operators acting on vectors . . . . . . . . . . . . . . . . . . . . . . 25
3.1.2 Vectors as operators acting on 1-forms . . . . . . . . . . . . . . . . . . . . . . 26
3.1.3 Tensors as operators acting on vectors and 1-forms . . . . . . . . . . . . . . . 27
3.1.4 Metric tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.1.5 Constructing higher rank tensors via outer multiplication of vectors and 1-forms 28
3.2 Bases and components of tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.1 Induced basis of 1-forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.2 Induced bases of tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2.3 Index notation of tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2.4 Coordinate bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.2.5 Coordinate components of df . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.2.6 Metric form and metric tensor . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.3 Basic tensor operations and tensor equations . . . . . . . . . . . . . . . . . . . . . . 30
3.4 Basis transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.4.1 Transformation of induced bases . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.4.2 Transformation of components . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3
4 CONTENTS

3.5 The operations of raising and lowering indexes of tensors . . . . . . . . . . . . . . . . 33


3.6 Symmetric and antisymmetric tensors . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.6.1 Symmetry with respect to a pair of indexes . . . . . . . . . . . . . . . . . . . 34
3.6.2 Antisymmetry with respect to a pair of indexes . . . . . . . . . . . . . . . . 34

4 Geometry of Riemannian manifolds 35


4.1 Parallel transport and Connection on metric manifolds . . . . . . . . . . . . . . . . . 35
4.1.1 Parallel transport of vectors. Connection . . . . . . . . . . . . . . . . . . . . 35
4.1.2 Connection of Euclidean space . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.1.3 Riemannian Connection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.2 Parallel transport of tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.2.1 Scalars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.2.2 1-forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.2.3 General tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.2.4 Metric tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.3 Absolute and covariant derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.3.1 Absolute and covariant derivatives of vector fields . . . . . . . . . . . . . . . 39
4.3.2 Absolute and covariant derivatives of 1-form fields . . . . . . . . . . . . . . . 40
4.3.3 Absolute and covariant derivatives of general tensor fields . . . . . . . . . . . 40
4.3.4 Absolute and covariant derivatives of scalar fields . . . . . . . . . . . . . . . . 41
4.3.5 General properties of covariant differentiation . . . . . . . . . . . . . . . . . . 41
4.3.6 The field of metric tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.4 Geodesics and parallel transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.5 Geodesic coordinates and Fermi coordinates . . . . . . . . . . . . . . . . . . . . . . . 42
4.5.1 Geodesic coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.5.2 Fermi coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.6 Riemann curvature tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.7 Properties of the Riemann curvature tensor . . . . . . . . . . . . . . . . . . . . . . . 47
4.8 Ricci tensor, curvature scalar and the Einstein tensor . . . . . . . . . . . . . . . . . . 48

5 Space and time in the theory of relativity 49


5.1 Physical Space and Time in Newtonian Physics . . . . . . . . . . . . . . . . . . . . . 49
5.2 Physical Space and Time in Special Relativity . . . . . . . . . . . . . . . . . . . . . . 50
5.3 Relativistic equations of motion of particle dynamics . . . . . . . . . . . . . . . . . . 52
5.4 Conservation laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.5 Relativistic continuity equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.6 Stress-energy-momentum tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.6.1 Energy-momentum vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.6.2 Stress-energy-momentum tensor of dust . . . . . . . . . . . . . . . . . . . . . 56
5.6.3 Energy-momentum conservation . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.6.4 Stress-energy-momentum tensor of perfect fluid . . . . . . . . . . . . . . . . 57
5.7 Space and Time in General Relativity . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.8 Einsteins equations of gravitational field . . . . . . . . . . . . . . . . . . . . . . . . 60
5.9 Newtonian limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

6 Schwarzschild Solution 67
6.1 Schwarzschild Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.1.1 Schwarzschild Solution in Schwarzschild coordinates . . . . . . . . . . . . . . 67
6.1.2 Schwarzschild Solution in Kerr coordinates . . . . . . . . . . . . . . . . . . . 69
6.1.3 Event horizon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
6.2 Gravitational redshift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
6.3 Integrals of motion of free test particles in Schwarzschild spacetime . . . . . . . . . . 72
6.4 Orbits of test particles in the Schwarzschild geometry . . . . . . . . . . . . . . . . . 75
6.5 Perihelion shift of planets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
CONTENTS 5

6.6 Bending of light . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

7 Appendix 85
7.1 Geometric units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
6 CONTENTS
Chapter 1

Introduction

Einsteins road to General Relativity began in November 1907. Two limitations of Special Relativ-
ity bothered him at that time1 . First, it applied only to uniform constant-velocity motion (inertial
frames). Second, it did not incorporate Newtons theory of gravity which conflicted with Special
Relativity as it assumed instantaneous interaction between distant objects whereas in Special Rela-
tivity no signal can propagate faster than the speed of light. I was sitting in a chair in the patent
office at Bern when all of a sudden a thought occurred to me, recalled Einstein. If a person falls
freely he will not feel his own weight. This simple observation hinted the deep connection between
gravity and accelerated frames and propelled Einstein on a eight-year effort to generalize his Special
Relativity. Like in the case of Special Relativity, the key physical ideas of the new theory, called
General Relativity, were developed by Einstein via thought experiments and below we describe
some of them .
First imagine a man in an enclosed chamber floating in deep space far removed from stars and
other appreciable mass. Thus, the gravitational force is very small and the man would experience
weightlessness. He must fasten himself with strings to one of the walls, otherwise he would risk to
fly to another wall as a result of a smallest impact (see figure 1.1).

Figure 1.1: Left panel: chamber in deep space. Right panel: chamber in free fall in Earths gravita-
tional field.
1 This introduction is based on Walter Isaacson, Einstein. His life and Universe, Simon & Schuster Paperbacks,

New York, 2008

7
8 CHAPTER 1. INTRODUCTION

Now imagine that the same chamber is released close to earth and now falls freely towards it,
accelerating all the time. Again, the man would experience weightlessness just like in deep space
(In fact, nowadays this is used to train astronauts). Indeed, as this was discovered by Galileo all
objects freely falling under the Earths gravity experience the same acceleration. As the result, the
man will not be pressed against the top or bottom walls of the chamber as the chamber accelerates
in exactly the same way as he does (see figure 1.1). Thus, we conclude that accelerated motion can
neutralize gravity and in this sense both phenomena are very similar.
Now consider the chamber resting on the ground. The normal reaction force of the ground
prevents the chamber from accelerated motion and the man as well as all objects are pressed against
the floor. Any object lifted from the floor and then released will fall back on the floor (see figure
1.2).
Next imagine this chamber in a deep space again but now a rope is attached to one of the walls
(the roof) and pulled up with a constant force (acceleration). The man inside the chamber observes
that he and all other objects freely floating inside the chamber before now begin to move with
exactly the same acceleration towards the opposite wall (the floor) just like observed by Galileo
in Pisa. Eventually, they all are pressed against the floor. Any object lifted from the floor and then
released falls back on the floor (see figure 1.2). All these observations naturally drive the man inside
to conclude that the chamber is in gravitational field. He might wonder why the chamber itself is
not in free fall in this gravitational field. Just then, however, he discovers the hook and the rope and
comes to the false conclusion that the chamber is suspended above the ground. This force, however
is not gravity. It is called inertial force but its effects are equivalent to the uniform gravity force.
Einstein called this the principle of equivalence: ... it follows that it is impossible to discover by
experiment whether a given system of coordinates is accelerated, or whether ... the observed effects
are due to a gravitational field.

a=g

Figure 1.2: Left panel: accelerated chamber in deep space. Right panel: chamber at rest on ground
in Earths gravitational field.

The fact that all bodies experience the same acceleration in gravitational field means that the
inertial mass equals (proportional) to the gravitational mass (charge). In the second law of Newton,

f = ma,

m is the inertial mass of a body. It describes bodys ability to resist the accelerating effect of force
9

f applied to it. In the Newtons law of gravity,

mg M
f = G r,
r3
mg is bodys gravitational mass. It describes the intensity of its gravitational interaction with
another body of gravitational mass M . Newton noticed that if

m = mg (1.1)

then the acceleration of the body is independent of its mass


M
a = G r.
r3
Thus, all bodies would experience exactly the same acceleration, just like discovered by Galileo.
Now we can see that Einsteins equivalence principle is rooted in the equivalence of inertial and
gravitational masses.
These experiments implied that that new relativistic theory of gravity could be constructed via
generalizing Special Relativity in such a way that it deals with not only inertial frames but also
accelerated frames. Special Relativity dismissed the notions of absolute space and time. It also
dismissed the notion of absolute motion, that is the motion in absolute space. Only the motion
relative to other physical bodies is considered as meaningful. This equally applies to the motion of
bodies and the motion of reference frames. However, only a particular kind of motion of reference
frames was considered in the original Special Relativity, namely the non-accelerated motion, and
hence only a particular kind of reference frames, namely the inertial frames. But why should some
reference frames be more special compared to others? If the absolute space does not exist and
thus only the relative motion is physically meaningful then whether the motion is accelerated or not
should also be relative. Similarly, there should not be a division on inertial and accelerated reference
frames and more general relativity theory should treat them equally. In particular, the physical laws
must have the same form (to be covariant) in all reference frames making no distinction between
inertial and accelerated ones. Hence the name of this new theory: General Relativity.

a=g

Figure 1.3: Left panel: Bending of light beam in the accelerated chamber in deep space. Right
panel: Bending of light beam in the chamber at rest on ground in Earths gravitational field.

Einstein also noted that principle of equivalence suggests bending of light rays in gravitational
field. Once again imagine a chamber that is accelerated in deep open space. Suppose that a laser is
10 CHAPTER 1. INTRODUCTION

mounted on one of the walls and sends a light beam perpendicular to the direction of acceleration.
Consider one of the emitted photons. At the moment of emission this photon moves perpendicular
to the direction of the chamber acceleration in the reference frame of the chamber. And so it does
in the inertial frame that is moving with the same velocity as the chamber at the time of emission.
Moreover in this inertial frame photons direction of propagation remains unchanged. However, by
the time it hits the opposite wall the chamber is already moving with finite speed relative to this
inertial frame. Thus, in the chamber frame the photon velocity must have a finite component along
the direction of effective gravity2 . This implies that in the chamber frame the photon trajectory is
bend (see fig.1.3). The equivalence principle then predicts exactly the same bending of light beams in
the gravitational field which produces gravitational acceleration of the same strength3 (see fig.1.3).
The conclusion that the light beams could be bend led to some interesting questions. If you think
about it is the light beams that are identified in practical geometry with straight lines builders
now use laser beams to mark off straight lines and built level houses. If a light beam curves in
gravitational field, how can a straight line be determined? One solution might be to liken the path
of a the beam to that of the line drawn on a sphere of on a surface that is warped. In such cases, the
shortest line between two points is a geodesic like an arc of a great circle on our globe. Perhaps, the
bending of light means that the fabric of space is curved by gravity and it can no longer be described
by Euclidean geometry. Moreover, since Special Relativity unites space and time into a single space-
time the new relativistic theory of gravity should rather consider warping of the space-time not just
space.
At first Einstein did not fully appreciate the power of mathematical formalism. When he was a
student of Zurich Polytechnic (1896-1900) he often skipped math classes and relied on notes taken
by his classmate and long-life friend Marcel Grossmann. His mathematical weakness was the reason
for not he but in fact his former math teacher in Zurich Polytechnic, Hermann Minkowski, who
made the key step in mathematical formulation of Special Relativity which lets the theory shine in
all its glory (and is used in all modern textbooks on the subject). His approach was the same one
suggested by the time traveler on the first page of H.G.Wells novel The Time Machine, published
in 1895: There are really four dimensions, three which we call the three planes of Space, and a
fourth, Time. Minkowski united space and time into a four-dimensional metric space, space-time.
He dramatically announced his new approach in lecture in 1908. The views of space and time which
I wish to lay before you have sprung from the soil of experimental physics, and therein lies their
strength, he said. They are radical. Henceforth space by itself, and time by itself, are doomed to
fade into mere shadows, and only a kind of union of the two will preserve an independent reality.
Characteristically for Einsteins views at that time he described Minkowskis work as superfluous
learnedness and joked, Since the mathematicians have grabbed hold of the theory of relativity, I
myself no longer understand it!
However, by 1920s he had become a convert to the faith in mathematical formalism because it
had proved so useful in his road to General Relativity. In 1912 he was desperate being unable to
generalize the relativity theory and wrote to Grossmann, Grossmann, youve got to help me or I
will go crazy. What he needed was help in finding a suitable mathematical language that would
express the new laws of gravity. Grossmann, after consulting the literature, recommended Einstein
the non-Euclidean geometry that has been devised by Bernhard Riemann (1826-1866) and tensor
calculus. Einstein took this on board and pursued a two-fisted approach. On the one hand, he
engaged in a physical strategy, in which he tried to build the theory from a set of requirements
dictated by his feel for the physics. On the other hand, he also pursued a mathematical strategy,
in which he tried to deduce the correct equations from the more formal math requirements using
the tensor calculus.
Using the mathematical strategy Einstein came very close to the final equations of General
Relativity already in 1912 (in The Zurich Notebook) but could not made the final step to make
them consistent with the requirement of energy and momentum conservation. So he turned more to
the physical strategy. It was a decision that he regretted later. The physical strategy did not work
for him. He lost valuable time and the final push to General Relativity turned into a race in which
2 See the aberration of light in the notes on Special Relativity
3 This effect was observed in 1919 and this was the first real test of General Relativity.
11

he almost had been overtaken by a brilliant mathematician, David Hilbert. Luckily for Einstein, he
returned to the mathematical strategy, just in time, and it proved spectacularly successfully. On
November 25, 1915 in his lecture The Field Equations of Gravitation, Einstein presented the final
result,
R 1/2g R = 8T ,
the equation that describes how matter tells space-time how to curve and the curved space-time
tells matter how to move.
In this course we will not follow all the steps of the complicated route that has lead Einstein to
General Relativity. Instead, and along with most modern textbooks, we will pursue the mathematical
strategy. This is the easiest and the shortest way indeed.
12 CHAPTER 1. INTRODUCTION
Chapter 2

From Euclidean space to surfaces


and metric manifolds

2.1 Metric form


2.1.1 The notion of metric form
Consider a plane in a 3-dimensional (3D) Euclidean space. This plane is a 2D Euclidean space.
Therefore, we can introduce Cartesian coordinates {x, y} for its points:

Figure 2.1:

If dl is the distance between infinitesimally close points (x, y) and (x + dx, y + dy) then

dl2 = dx2 + dy 2 . (2.1)


- this is the metric form of the plane in Cartesian coordinates {x, y}. We may introduce new
coordinates {x1 , x2 } via e.g.

x1 = x y, x2 = x 2y. (2.2)
1 2
What is dl in terms of dx and dx ? From eq.(2.2) one has
x = 2x1 x2 , y = x1 x2 ,
then
dx = 2dx1 dx2 , dy = dx1 dx2 ,

13
14CHAPTER 2. FROM EUCLIDEAN SPACE TO SURFACES AND METRIC MANIFOLDS

and finally
dl2 = dx2 + dy 2 = 5(dx1 )2 6dx1 dx2 + 2(dx2 )2
or

dl2 = 5(dx1 )2 3dx1 dx2 3dx2 dx1 + 2(dx2 )2 . (2.3)


We may write this as
2 X
X 2
dl2 = gij dxi dxj (2.4)
i=1 j=1
where
g11 = 5, g12 = g21 = 3, g22 = 2.
For any choice of coordinates the metric form can be written as in eq.(2.4) with gij = gji . For
example, if x1 and x2 were Cartesian coordinates (just like x and y) then we would have
g11 = 1, g12 = g21 = 0, g22 = 1.
If instead of a 2D Euclidean plane we consider an n-dimensional Euclidean space then we obtain
a similar result: the distance between its two infinitesimally close points can be written as
n X
X n
dl2 = gij dxi dxj (2.5)
i=1 j=1

where
gij = gji
i
for any set of coordinates {x }, i = 1, 2, ..., n. Coefficients gij of the metric form are often shown as
components of a n n matrix. For example, in the case (2.3)
 
5 3
gij = ,
3 2
and in the case (2.1)  
1 0
gij = ,
0 1
where it is assumed that x1 = x and x2 = y.

Einstein summation rule:


Any index appearing once as a lower index and once as an upper index of the same indexed object
or in the product of a number of indexed objects stands for summation over this index. Such index
is called a dummy index. Indexes which are not dummy are called free indexes.

According to this rule we can rewrite eq.(2.5) in a more concise form:


dl2 = gij dxi dxj . (2.6)
This rule allows to simplify expressions involving multiple summations. Here are some more exam-
ples:
Pn
1. ai bi stands for i=1 ai bi ; here i is a dummy index;
2. ai bi stands for a product of ai and bi where i can have any value between 1 and n; here i is a
free index.
Pn
3. ai bkij stands for i=1 ai bkij ; here k and j are free indexes and i is a dummy index;
f Pn i f
4. ai x i stands for i=1 a xi ; thus, index i in the partial derivative xi is treated as a lower
index;
2.1. METRIC FORM 15

2.1.2 Metric forms of surfaces:


For any smooth surface in Euclidean space the distance between its any two infinitesimally close
points can be found in terms of coordinates introduced on the surface. For example, consider a
sphere of radius r in 3D Euclidean space. This is a 2D surface and one needs two coordinates to
mark its points. Introduce the usual spherical coordinates {, }.

Figure 2.2:

Then for the Cartesian coordinates {x, y, z} shown in the figure



x = r sin cos ,
y = r sin sin , .
z = r cos

This gives us
dx = r cos cos d r sin sin d,
dy = r cos sin d + r sin cos d,
dz = r sin d

and
dl2 = dx2 + dy 2 + dz 2 = ... = r2 d2 + r2 sin2 d2 . (2.7)
Thus,

r2
 
0
gij =
0 r sin2
2

where we assume that x1 = and x2 = .

Locally Cartesian coordinates:

It is impossible to introduce Cartesian coordinates for the whole sphere, that is such two coor-
dinates x1 and x2 that
dl2 = (dx1 )2 + (dx2 )2
everywhere on the sphere (a sphere is not like a plane). However, there exist so-called locally
Cartesian coordinates.
16CHAPTER 2. FROM EUCLIDEAN SPACE TO SURFACES AND METRIC MANIFOLDS

Take some point of the sphere, denote it as A. Suppose its spherical coordinates are a and a .
Near A introduce new coordinates
 1
x = r( a )
.
x2 = r sin a ( a )
Then
dx1

= rd
,
dx2 = r sin a d
and
dx1 /r

d =
.
d = dx2 /r sin a
Substitute these into eq.(2.7) to obtain the metric form
 2
2 1 2 sin
dl = (dx ) + (dx2 )2 .
sin a
At the point A this becomes
dl2 = (dx1 )2 + (dx2 )2 .
Thus, near point A the metric form is the same as the metric form of a 2D Euclidean space with
Cartesian coordinates {xi }. Because of this property, the sphere is called locally Euclidean or
Riemannian. (All smooth surfaces in Euclidean space are locally Euclidean.)

2.1.3 Lengths of curves


Let {xi } be some arbitrary (curvilinear) coordinates in n-dimensional Euclidean space and xi = xi ()
be a curve in the space. ( is the curve parameter; one can view it as a coordinate introduced only
for the points of the curve).

Figure 2.3:

The length of the curve between its any two points, A and B, is given by
ZB ZB ZB 1/2
i j 1/2 dxi dxj
l = dl = (gij dx dx ) = gij d. (2.8)
d d
A A A

2.1.4 Coordinate transformations:


0 0
Introduce arbitrary new coordinates {xi } whose coordinate lines may be curved. xi are functions
of the old coordinates xk : 0 0
xi = xi (xk ).
0
Inversely, xk are functions of xi : 0
xk = xk (xi ).
2.2. VECTORS, BASES, AND COMPONENTS OF VECTORS 17

Then
0 0
dl2 = gij dxi dxj = gi0 j 0 dxi dxj , (2.9)
where
0 0
xl xm xl xm
gij = gl 0 m0 , and g i 0j0 = glm . (2.10)
xi xj xi0 xj 0
Eq.(2.10) is the transformation law for the components of the metric form.
If {xi } are Cartesian then 
1 if l = m
glm = (2.11)
0 if l 6= m
and the second equation in eq.(2.10) reduces to
n
X xl xl
gi0 j 0 = . (2.12)
xi0 xj 0
l=1

2.2 Vectors, bases, and components of vectors


In Euclidean geometry vectors are defined as straight arrows. The magnitude of a vector is the
length of the arrow. We denote it as |a|.

2.2.1 Coordinate bases


Let {xi } be Cartesian coordinates of n-dimensional Euclidean space. Let ei be the unit vectors
pointing in the direction of the xi -coordinate axis. The set of all n vectors ei at any point of the
space forms a vector basis at this point, the Cartesian basis. If

a = ai ei
then ai are the components of a in this basis. Vector

r = xk ek . (2.13)

whose base coincides with the origin of the coordinate system and whose tip coincides with the point
with coordinates xk is called the position vector of this point.
0 0
Introduce arbitrary new coordinates {xi } whose coordinate lines may be curved. xi are functions
of the old Cartesian coordinates xk : 0 0
xi = xi (xk ).
0
Inversely, xk are functions of xi : 0
xk = xk (xi ).
Definition: The set of vectors 0
ei0 = r/xi (2.14)
defined at the point with position vector r provides us with a basis which is called the coordinate
0 0
basis of the {xi } coordinates at this point. ei0 is tangent to the xi coordinate line passing through
this point.

Comment: {ek } is the coordinate basis of original Cartesian coordinates, that is


r
ej = .
xj
0 0
If a = ai ei0 then ai are the components of a in the basis {ei0 }.
18CHAPTER 2. FROM EUCLIDEAN SPACE TO SURFACES AND METRIC MANIFOLDS

Figure 2.4:

2.2.2 Coordinate transformations


0
Consider transformation from coordinates {xi } to coordinates {xi }, both being arbitrary curvilinear
coordinates. Then
0
xk xi
e i0 = e , and ek = e 0. (2.15)
xi0 k xk i
If
0
a = ai ei = ai ei0
then 0
k xk i0 xi ki0
a = a , and a = a . (2.16)
xi0 xk

Definition: Kroneckers delta: 


1 if k = j
jk = (2.17)
0 6 j
if k =

2.3 Metric form and the scalar product


If ai and bi are the Cartesian components of vectors a and b then
n
X
ab= ai bi . (2.18)
i=1

n
X
|a|2 = a a = (ai )2 . (2.19)
i=1

The first equation can also be written as

a b = gij ai bj . (2.20)

where gij are the Cartesian components of the metric form (see eq.2.11).
2.4. GEODESICS AND THE VARIATIONAL PRINCIPLE 19

0 0
In fact, if ai and bi are the components of a and b and gi0 j 0 are the components of the metric
form in the coordinate basis of any other coordinate system we still have
0 0
a b = gi0 j 0 ai bj . (2.21)

Thus, expression (2.20) for the scalar product of two vectors is invariant under coordinate transfor-
mations(!)
If gij are the components of the metric form in some coordinate system and {ei } is the coordinate
basis of this system then

gij = ei ej . (2.22)
Consider an infinitesimally small vector dx connecting points with coordinates xi and xi + dxi .
The components of dx in the coordinate basis are dxi . The magnitude of dx is the distance dl
between the points. Then from the invariant expression eq.(2.21) one has

dl2 = dx dx = gij dxi dxj (2.23)


in agreement with eq.(2.6)

2.4 Geodesics and the variational principle


2.4.1 Euler-Lagrange Theorem
Consider the functional

ZB
lAB = L(xk , xk )d (2.24)
A

where x = x () (k = 1, 2, ..., n) are functions of and xk = dxk /d.


k k

The functions which extremise lAB and satisfy the boundary conditions

xk (A ) = xkA , xk (B ) = xkB (2.25)

are solutions of the following ODEs


d L L
=0 (k = 1, 2, ..., n) (2.26)
d xk xk
These ODEs are known as Euler-Lagrange equations.

2.4.2 Geodesics
Consider an n-dimensional Euclidean space or even a smooth surface in a higher dimensional Eu-
clidean space. Let xi , i = 1, 2, ..., n be some arbitrary coordinates in this space or surface and gij
are the corresponding components of the metric form.
Consider a curve xk = xk () connecting points A and B with coordinates xkA and xkB , that is

xk (A ) = xkA , xk (B ) = xkB .
20CHAPTER 2. FROM EUCLIDEAN SPACE TO SURFACES AND METRIC MANIFOLDS

The distance between A and B along this curve is

ZB
lAB = L(xk , xk )d
A

where Lagrangian L is
L(xk , xk ) = [gij xi xj ]1/2 . (2.27)
k
Note that in general gij = gij (x ).

Definition: Curves that extremise distances between all its points are called geodesics.

From the Euler-Lagrange theorem it follows that geodesics are solutions of the Euler-Lagrange
equations with Lagrangian (2.27). Instead of the Lagrangian (2.27) one can also use the Lagrangian

L(xk , xk ) = gij xi xj . (2.28)

This will result in the same curves but with different parametrization. Namely, will be a normal
parameter, that is such a parameter that

d = adl,

where a =const and l is the length of the geodesic (as measured from an arbitrary point of the
geodesic).

2.4.3 Examples of geodesics:


Euclidean space:

If xk are Cartesian coordinates then the Lagrangian (2.28) reads


n
X
L= (xk )2
i=k

and the corresponding Euler-Lagrange equations reduce to

dxk
= 0,
d
The solutions of these equations,
xk () = ak + bk ,
describe straight lines.

2D sphere in a 3D Euclidean space:

Consider a sphere of radius r with spherical coordinates {, }. Then the Lagrangian (2.28)
reads
L = r2 (2 + sin2 2 )
2.4. GEODESICS AND THE VARIATIONAL PRINCIPLE 21

and the corresponding Euler-Lagrange equations reduce to


 
d 2 d
d sin d = 0
 2
d d d

d d sin cos d = 0.

It is easy to verify that functions

() = a, () = b

deliver particular solutions to these equations. They describe meridians of the sphere. Each
meridian is a great circle, that is a circle formed by the intersection of the sphere and a plane
passing through its center. All other geodesics of the sphere are also great circles.

Figure 2.5: Meridian - an example of a great circle


22CHAPTER 2. FROM EUCLIDEAN SPACE TO SURFACES AND METRIC MANIFOLDS

2.5 Non-Euclidean geometry of a Euclidean sphere


Geodesics is a generalization of straight lines. Using geodesics one can build various geometrical
constructions on surfaces analogous to those of Euclidean spaces e.g. circles, triangles, rectangles
etc. They will have somewhat different geometrical properties.
Consider a 2D sphere in a 3D Euclidean space. In contrast to a 2D Euclidean space one finds
the following properties:

Figure 2.6: Geodesics of the sphere are closed curves

Figure 2.7: Different geodesics intersect at more than one point.


2.6. MANIFOLDS 23

Figure 2.8: The sum of angles of a triangle exceeds 2;

Figure 2.9: The circumference of a circle of radius R is l = 2r sin(R/r) < 2R.

2.6 Manifolds
Definition. A set of points, M, is called an n-dimensional manifold if any point of M has a
neighbourhood that allows one-to-one continuous map onto an open set in Rn (n-dimensional real
space). In other words one can introduce n continuous coordinates at least locally.

Definition. A n-dimensional manifold, M, is called a space if there exists a one-to-one continuous


map of the whole of M onto the whole of Rn . In other words one can introduce n continuous
coordinates globally.

Definition. When a manifold is attributed with distance between its points, via a metric form
(metric tensor), it is called a metric manifold.

Definition. A metric manifold is called Riemannian (or locally Euclidean) if for its every point
there exist local coordinates such that the metric form at this point has the components

1 if l = m;
glm = (2.29)
0 if l 6= m.

Such coordinates are called locally Cartesian.


Like in the case of the sphere considered in the previous section one can use geodesics to build
various geometrical constructions on Riemannian manifolds, and their properties may well be very
24CHAPTER 2. FROM EUCLIDEAN SPACE TO SURFACES AND METRIC MANIFOLDS

different from those in Euclidean geometry.

Definition. A Riemannian manifold is called a Euclidean space if there exist global coordinates,
called Cartesian, such that the metric form has components (2.29) at every point of the manifold.

A 2-dimensional sphere in a 3-dimensional Euclidean space is a 2-dimensional Riemannian manifold


but not a Euclidean space. A 2-dimensional plane in a 3-dimensional Euclidean space is a 2-
dimensional Euclidean space. All smooth surfaces in a Euclidean space are Riemannian manifolds.
A manifold is not necessarily a surface in a Euclidean or any other space. The spacetime of
General Relativity is an example of such manifold.

2.7 Vectors as operators


2.7.1 Basic idea
Vectors defined as straight arrows do not suit surfaces and manifolds. Such straight arrows cannot
belong to curved surfaces and, at most, can only be tangent to them, unless they are infinitesimally
small.
Vectors defined as directed bits of surface geodesics do not allow to introduce meaningful opera-
tions of addition and multiplication by real number, unless their are infinitesimally small and, thus,
indistinguishable from straight arrows tangent to the surface.
Cartan proposed to define vectors as directional derivatives. Consider a n-dimensional surface
with coordinates {xi } and a particle moving over the surface. The particle coordinates are functions
of time:
xi = xi (t). (2.30)
These equations describe a curve on the surface, the particle trajectory. t plays the role of its
parameter. The derivatives
dxi
vi = (2.31)
dt
have the meaning of velocity components. Consider the differential operator

d
= vi i , (2.32)
dt x

called the directional derivative along the curve (2.30). Note that v i are components of the operator
d/dt in the basis of partial derivatives /xi . Hence, the idea to identify the velocity vector with
this directional derivative and treat the partial derivatives and its local coordinate basis:

d
v= , ei = . (2.33)
dt xi
Then eq.(2.32) reads
v = v i ei , (2.34)

Addition and multiplication of Cartans vectors is defined via eq.2.32. That is

c = a + b if ci = ai + bi ;
and
a = b if ai = bi .
The set of all vectors defined this way at any particular point of the surface form an n-dimensional
vector space associated with this point.
2.7. VECTORS AS OPERATORS 25

2.7.2 Coordinate transformations


0
Introduce new coordinates, {xi }. According to the chain rule:
0
xk xi
0 = and = .
x i xi0 xk x k xk xi0
or,
0
xk xi
e i0 = ek , and e k = e 0,
xi 0
xk i
exactly as in eq.(2.15). Then from
0
v = v i e i = v i e i0 .
one has
0
xk i0 i0 xi k
vk = v , and v = v , (2.35)
xi0 xk
just like in eq.(2.16). Thus, the new definition of vectors as operators leads to the same transforma-
tion laws for their components as before. Note that neither the trajectory nor its parameter t are
effected by such transformation. Thus, the directional derivative v = d/dt is not effected either. It
is completely independent on the choice coordinates and exists even if no coordinates are introduced
altogether.

2.7.3 Magnitudes of vectors and the scalar product


Let v and w be two Cartan vectors (operators). Because of the transformation law (2.35) the quantity
gij v i wj remains invariant under coordinate transformations (such quantities are called scalars.) This
can still be called the scalar product of v and w

v w = gij v i wj .

Moreover, gij v i v j , provides meaningful definition for the magnitude v = |v| of vector v:

|v|2 = gij v i v j .

Indeed, consider the infinitesimal displacement vector of our particle,

dx = vdt.

Its components
dxi = v i dt
are the differences in coordinates of the two points on the particle trajectory separated by time dt.
The distance between these points point is given by

dl2 = gij dxi dxj = (gij v i v j )dt2 = v 2 dt2 .

Thus, we have
dl = vdt
as usual.

In fact, none of the properties of Euclidean vectors introduced as arrows is lost by Cartans
vectors introduced as operators.
26CHAPTER 2. FROM EUCLIDEAN SPACE TO SURFACES AND METRIC MANIFOLDS
Chapter 3

Tensors

Tensors are used not only in the Theory of Relativity but also in many fields of Newtonian physics,
sometimes without proper introduction.

3.1 Tensors as operators


Consider an n-dimensional manifold. Let P be a point of the manifold. Denote as Tp the set of all
vectors defined at P. Tp is an n-dimensional vector space (see Sec.1.7.1)

3.1.1 1-forms as operators acting on vectors


Definition. A 1-form q defined at P is a linear scalar operator acting on vectors from Tp . That is
1. q : Tp R;
2. For any v, u Tp and a, b R

q(av + bu) = aq(v) + bq(u). (3.1)

The set of all 1-forms defined at P is denoted as Tp . This is an n-dimensional vector space with

1. Zero-element 0 such that


0(u) = 0 for any u Tp ;

2. Operation of addition:
q = p + w if for any u Tp
q(u) = p(u) + w(u); (3.2)

3. Operation of multiplication:
q = ap if for any u Tp
q(u) = ap(u). (3.3)

To stress that 1-form q is an operator it is often shown as

q( )

where the space inside the brackets is a slot to be filled with a vector.

27
28 CHAPTER 3. TENSORS

Examples of 1-forms:

To any vector v Tp there corresponds a 1-form v introduced via the scalar product operation
as follows
v(u) = v u for any u Tp . (3.4)
This 1-form is called dual to the vector v. The condition (3.1) is satisfied because

v (au + bw) = a(v u) + b(v w).

Gradient of a scalar function.


such that for any infinites-
Let f be a scalar function defined on the manifold. The 1-form df
imally small vector dx from Tp
(dx) = df
df (3.5)
is called the gradient of f at point P.

Figure 3.1:

Since df = (f /xi )dxi we have


(dx) = f dxi .
df
xi
This suggests that for any vector u

(u) = f ui .
df (3.6)
xi
The expression on the right is indeed a scalar (a number which the same for all coordinates):
 i
f xi
  
f i f x j 0 0 f j 0
u = u = uj = u .
xi xi xj 0 xi xj 0 xj 0

3.1.2 Vectors as operators acting on 1-forms


One can associate with any vector u Tp a linear scalar operator acting on 1-forms from Tp via

u(q) = q(u) < u, q > (3.7)


From eqs.(3.2) and (3.3) it follows that indeed

u(ap + bq) = au(p) + bu(q).

To stress this role of vectors they are often shown like

u( )

where the space inside the brackets is a slot to be filled with a 1-form.
3.1. TENSORS AS OPERATORS 29

3.1.3 Tensors as operators acting on vectors and 1-forms



Definition. An lm -type tensor defined at point P is a linear scalar operator with l slots for 1-forms
from Tp and m slots for vectors from Tp . Such tensor can also be called as l-times contravariant
and m-times covariant. The total number of slots, r = l + m, is called the rank of the tensor.

Thus,
1

1. Any vector is a -type tensor;
0

2. Any 1-form is a 01 -type tensor;

3. If, for example, M ( , ) is 11 -type tensor with the first slot reserved for 1-forms then

M (q, u) R;

M (ap + bq, u) = aM (p, u) + bM (q, u);

M (p, au + bv) = aM (p, u) + bM (p, v);
l

The set of all m -type tensors defined at point P is an nr -dimensional vector space with
1. Zero element O such that
O(u, . . . , q) = 0 for any l vectors from Tp and m 1-forms from Tp ;

2. Operation of addition
S =T +K if for any l vectors from Tp and m 1-forms from Tp

S(u, . . . , q) = T (u, . . . , q) + K(u, . . . , q); (3.8)


3. Operation of multiplication by real numbers
S = aT, where a R, if for any l vectors from Tp and m 1-forms from Tp

S(u, . . . , q) = aT (u, . . . , q); (3.9)

3.1.4 Metric tensor


0

Definition. A 2 -type tensor g( , ) such that for any two vectors v, u Tp
g(v, u) = v u (3.10)
is call the metric tensor.

Notice that the metric tensor and the one-form v dual to the vector v (see Sec.2.1.1) are related
via

v( ) = g(v, ). (3.11)
Indeed, this ensures that
v(u) = g(v, u) = v u.
Later on we will describe a relationship between the metric tensor and the metric form
30 CHAPTER 3. TENSORS

3.1.5 Constructing higher rank tensors via outer multiplication of vectors


and 1-forms
The following examples explain the operation of outer multiplication. is the symbol of this
operation. Here v, u etc. are vectors from Tp , and p, q etc. are 1-forms from Tp .

Example
F ( , ) = u( ) v( )
2

is a 0 -type tensor such that for any p, q

F (p, q) = u(p)v(q);

Example
S( , ) = q( ) v( )
1

is a 1 -type tensor such that for any p, u

S(u, p) = q(u)v(p);

Example
D( , , ) = q( ) v( ) t( )
1

is a 2 -type tensor such that for any p, u, s

D(u, p, s) = q(u)v(p)t(s);

etc.

3.2 Bases and components of tensors


3.2.1 Induced basis of 1-forms
Let {ei }ni=1 be a basis in Tp . Then for any u Tp

u( ) = ui ei ( ). (3.12)
ui are the components of u in this basis. Note that i is an upper index.
Let {wi }ni=1 be a basis in Tp . Then for any q Tp

q( ) = qi wi ( ), (3.13)
where qi are the components of q in this basis. Note that i is a lower index in qi . This is to make
clear that we are dealing with the components of a 1-form but not a vector. In order to utilise the
Einstein summation rule in equations like eq.(3.13) we are then forced to use upper indices for the
basis 1-forms wi .
From eqs.(3.12,3.13) one has

wi (u) = uj wi (ej );
ei (q) = qj ei (wj ); (3.14)
q(u) = qi uj wi (ej ).

Definition. The basis {wi } is called induced by the basis {ei } if

wi (ej ) = ji . (3.15)
3.2. BASES AND COMPONENTS OF TENSORS 31

Then eqs.(3.14) simplify so that we have

wi (u) = ui ;
ei (q) = qi ; (3.16)
q(u) = qi ui .
Such simplifications is the main reason for using induced bases of 1-forms.

3.2.2 Induced bases of tensors


Induced bases of tensors are introduced for the same reason (simplicity). The following examples
explain how these bases are constructed:

(a) The induced basis of 11 -type tensors with the first slot intended for 1-forms is {ei wj },
where {wi } is the induced basis of 1-forms.

If F ( , ) is such a tensor and F ij are its components in this basis, then

F ( , ) = F ij ei ( ) wj ( ); (3.17)

F ij = F (wi , ej ); (3.18)

F (q, u) = F ij qi uj . (3.19)


(b) The induced basis of 02 -type tensors is {wi wj }. If g( , ) is such a tensor and gij are its
components in this basis then

g( , ) = gij wi ( ) wj ( ); (3.20)

gij = g(ei , ej ); (3.21)

g(u, v) = gij ui v j . (3.22)


etc.

3.2.3 Index notation of tensors


The number and position of indexes of tensor components reveal all the general information about
tensors as operators. For example if tensor T has components T ij kl this immediately tells us that

1. T is a 4th rank tensor;



2. T is a 22 -type tensor;

3. Its 1st and 3rd slots are for 1-forms whereas its 2nd and 4th slots are for vectors. That is

T ( , , , ) = T ij kl ei ( ) wj ( ) ek ( ) wl ( )

Because of this nice property it is a custom to introduce tensors simply by showing their components.
Hence, it is perfectly OK to say

Let us consider tensor T ij kl


32 CHAPTER 3. TENSORS

3.2.4 Coordinate bases


In Section 1.7.1 we introduced the coordinate basis
{/xi } i = 1, . . . , n
of vectors ( {r/xi } in the old fashion notation; see also Sec.1.2.1). The corresponding induced
bases of other tensors are also called coordinate. The coordinate basis of 1-forms is denoted as
i
} i = 1, . . . , n.
{dx
0

The coordinate basis of 2 -type tensors is then

i dx
{dx j } i, j = 1, . . . , n

etc.

3.2.5
Coordinate components of df
is the gradient of the scalar function f then
If df

= f dx
df i. (3.23)
xi

in the coordinate basis of the coordinates {xi }.


Thus, f /xi are the components of df

3.2.6 Metric form and metric tensor


The metric form

dl2 = gij dxi dxj (3.24)


i i i
gives us the distance, dl, between the point x and the point x + dx . Consider the infinitesimally
small vector

dx = dxi i .
x
If g( , ) is the metric tensor than

dx dx = g(dx, dx) = gij dxi dxj , (3.25)


where  

gij = g ,
xi xj
are the coordinate components of the metric tensor. Comparison of eq.(3.24) with eq.(3.25) shows
that
the components gij of the metric form are nothing else but the components of the metric tensor
in the coordinate basis.

3.3 Basic tensor operations and tensor equations


Definition. Operations with tensors which produce other tensors are call tensor operations.

All such operations can be introduced without making use of bases and components of tensors.
However, in this section we only describe the effect they have on components of tensors. In fact,
this is a very concise and fully comprehensive way of describing tensor operations. Keep in mind
that what is shown below are just examples involving tensors of particular types. Generalisation,
however, is very straightforward.
3.3. BASIC TENSOR OPERATIONS AND TENSOR EQUATIONS 33

1. Addition:
C ij = Ai j + B ij (3.26)

when tensor C is a sum of tensors A and B;

2. Multiplication by a real number:


C ijk = aAi jk (3.27)

when tensor C is a product of real number a and tensor A;

3. Outer multiplication:
T ijkl = Dij Bkl (3.28)

when T is the outer product of D and B (T = D B);

4. Contraction of a single tensor:


S ij = T illj (3.29)

when S is the result of contracting T over its second upper and first lower indexes (l is a
dummy index);

5. Contraction of two tensors:


T ij = Dil Blj (3.30)

when T is the result of contraction D and B over the 2nd upper index of D and the first lower
index of B (l is a dummy index);

Equations relating different tensors by means of tensor operations are called tensor equations.
Thus, equations 3.26-3.30 are examples of tensor equations. All tensor equations satisfy the following
simple formal rules:

1. All terms of tensor equations must have the same number and positions of free
indexes. Thus, for example, if i is an upper free index in one of the terms then it must be
an upper free index in all other terms.

Examples:
S ij = T ikj + P ij

is not a proper tensor equation whereas

S ij = T ikkj + P ij

is.

2. The order of free indexes is not important, so

S ij = T ikkj + Dj i

is still OK.

3. Also remember not to write a lower index just below an upper index because this makes the
order of slots ambiguous. That is
Sji = Tkj
ik
+ Pji

is not OK.
34 CHAPTER 3. TENSORS

Theorem

If a tensor equation involves m indexed objects and we know that


m 1 of them are tensors then the remaining one is also a tensor. (3.31)

For example, if T ikl and ui are tensors and

T ikl = ui Bkl

then Bkl is also a tensor. This theorem is proved using the transformation law of components of
tensors.

3.4 Basis transformation


Here we study the way the components of tensors transform as the result of the transformation
of the vector basis {ei } and, hence, the transformations of the induced bases of all other tensors
triggered by this transformation of the vector basis. The old style definition of tensors was based
on this transformation law. As before, we shell use dash to indicate new bases and components
of tensors.
Any vector of the new vector basis is a linear combination of the vectors of the old vector basis.
Hence,

ek0 = Aik0 ei . (3.32)

Aik0 is the transformation matrix (not a tensor). Similarly,


0
ek = Aik ei0 . (3.33)
0
The transformation matrix Aik is inverse to Aik0 , that is
0 0 0
Aik Akj0 = ji 0 and Aik Aji0 = kj . (3.34)
0

If {ei = xi } and {ei0 = xi0
} are the coordinate bases of coordinates {xi } and {xi } respectively
then
0
0 xi xj
Aik = and Aji0 = . (3.35)
xk xi0

3.4.1 Transformation of induced bases


The corresponding transformation of the induced basis of 1-forms is
0 0 0
wk = Aki wi and wk = Aki0 wi . (3.36)

Given new bases of vectors and 1-forms one can construct induced bases of all higher rank tensors.
For example
0 0
ei0 wj = Aki0 Ajl ek wl ; (3.37)

ei0 ej 0 = Aki0 Alj 0 ek el ; (3.38)


0 0 0 0
wi wj = Aik Ajl wk wl . (3.39)
3.5. THE OPERATIONS OF RAISING AND LOWERING INDEXES OF TENSORS 35

3.4.2 Transformation of components


Given the new basis (new induced basis) one can find the components of any tensor in this new basis
and relate them to the original components in the old basis.

Vectors:
0 0 0
ui = Aik0 uk and ui = Aik uk ; (3.40)

1-forms:
0
qi = Aki qk0 and qi0 = Aki0 qk ; (3.41)

Higher rank tensors:

General rule: Each upper index is treated as a vector index


and each lower index as an index of a 1-form. (3.42)

For example,

0 0 0 0
T ij 0 = Aik Alj 0 T kl and T ij = Aik0 Alj T kl0 (3.43)

3.5 The operations of raising and lowering indexes of tensors


ji can be considered as a tensor because it satisfies the tensor transformation law, eq.3.42. Indeed,
0 0 0
Aik0 Alj lk0 = Ail0 Alj = ji .

Now suppose we are dealing with a metric manifold. Consider the tensor equation

g ij gjk = ki , (3.44)

where gjk is the metric tensor. Since both gjk and kj are tensors so must be g ij (see eq.3.31). This
tensor is also called the metric tensor. This makes perfect sense because g ij is uniquely defined by
gij .
We already know that the metric tensor allows to relate vectors and 1-forms (see eq.3.11):

u( ) = g(u, ).

In components this reads

ui = gij uj . (3.45)
Given eq.3.44 we invert eq.3.45 to find

ui = g ij uj . (3.46)
Thus, the metric tensor allows to define a one-to-one relationship (map) between vectors
and 1-forms. Now we can interpret u as a first rank tensor which can be represented
either as a vector, u, or a 1-form, u.
Often, the components ui are called the covariant components of u and ui the contravariant
components of u. This is because the transformation law for ui is the same as the one for the basis
vectors (they covary) and it is different for ui :
0 0
ui0 = Aji0 uj and ei0 = Aji0 ej but ui = Aij uj .
36 CHAPTER 3. TENSORS

Similarly, the metric tensor is used to unify all tensors of the same rank. For example, if
T ij = gjk T ik
Ti j = gik T kj (3.47)
Tij = gik gjl T kl
then T ij , T ij , Tij , and Ti j are different representations of the same tensor T . For this reason
the operations like (3.45-3.47) are called rising and lowering indexes of a tensor.

In the operation of contraction it does not matter which of the dummy indexes is lower and
which is upper. For example,
T ik uk = T ik uk . (3.48)
The vector-gradient of a scalar function f , f , if defined as
f
i f = g ij dfj = g ij
. (3.49)
xj
and f represent the same 1st rank tensor called the gradient of f .
Thus, df

3.6 Symmetric and antisymmetric tensors


In many applications we deal with so-called symmetric and antisymmetric tensors. Here we explain
what they are by example.

3.6.1 Symmetry with respect to a pair of indexes


Tensor T ijk is called symmetric with respect to i and j if
T ijk = T jik . (3.50)
When any of the indexes of T is lowered the symmetry is preserved. That is
T ijk = T jik
T ijk = Tj ik (3.51)
Tij k = Tji k .

3.6.2 Antisymmetry with respect to a pair of indexes


Tensor T ijk is called antisymmetric with respect to i and j if
T ijk = T jik . (3.52)
When any of the indexes of T is lowered the symmetry is preserved. That is
T ijk = T jik
T ijk = Tj ik (3.53)
Tij k = Tji k .
It is easy to show that if T ijk is symmetric with respect to i and j and Fij is antisymmetric with
respect to i and j then

T ijk Fij = 0. (3.54)


Indeed,
T ijk Fij = T ijk Fji = T jik Fji = T ijk Fij ,
where the first equality is due to the antisymmetry of Fij , the second one is due to symmetry of
T ijk and the third one stands because we simply rename dummy indexes. Next we obtain
2T ijk Fij = 0,
from which eq.3.54 follows.
Chapter 4

Geometry of Riemannian manifolds

4.1 Parallel transport and Connection on metric manifolds


In the previous chapter we discussed operations on tensors defined at a single point of a manifold.
By means of such operations we can compare two tensors defined at the same point and measure
the difference between them. Tensors are here to describe objects of real life and in real life there
are meaningful ways of comparing similar objects at different spatial locations. One way is to bring
the objects to the same location so that direct comparison is possible. In this approach we have to
make sure that the transported objects are not modified along way. This can be done via control
measurements carried out using standard tools. In geometry such transport of tensors from one
to another point of a manifold is called the parallel transport and the control measurements are
introduced by means of the metric tensor. Indeed, the metric tensor is a mathematical object which
allows us to introduce the very basic and hence the most important physical measurements, the
measurements of length (and time as we shell see later). In order to agree with its description as a
standard control tool the metric tensor must be the same, in some absolute sense, everywhere on the
manifold. That is its parallel transport from point A to point B should give us exactly the metric
tensor already defined at point B. In other words defining the metric tensor on a manifold should
be consistent with 1) its defining at some single point of the manifold and 2) its parallel transport
to all other points.

4.1.1 Parallel transport of vectors. Connection


Consider vector a at point P of a metric manifold. Parallel transport it along the displacement
vector dx into the infinitesimally close point S (assuming that there is a meaningful way of such
transport.) Denote the result as a. This operation can be expressed as

a = (P, a, dx), (4.1)

where is the operator of parallel transport. It is also called the connection. Once this operator
is introduced at every point of the manifold we have means of parallel transporting vectors (and
tensors as well). Notice that is not a tensor as equation (4.1) involves vectors defined at different(!)
points of the manifold.

37
38 CHAPTER 4. GEOMETRY OF RIEMANNIAN MANIFOLDS

Basic requirements on parallel transport:

1.
If a = 0 then a = 0; (4.2)

2.
If dx = 0 then a = a; (4.3)

3. Linearity 1.
If a = b + c then a = b + c. (4.4)

4. Linearity 2. Introduce local coordinates {xi } on the manifold. Let ai and ai be the compo-
nents of a and a in the coordinate bases at P and P respectively.

If i
ai ai = da for dx(1) i
then ai ai = da for dx(2) = dx(1) . (4.5)

It is easy to see that these requirements are satisfied only if

ai = ai i jk aj dxk (4.6)

where i jk are called the coordinate components of . They are also known as Christoffels symbols
of the first kind.

4.1.2 Connection of Euclidean space


In Euclidean space the parallel transport of tensors amounts to keeping their Cartesian components
fixed (by definition). Thus, in Cartesian coordinates {xi } we must have

i jk = 0. (4.7)

If was a tensor than eq.(4.7) would hold in any coordinates, but it is not. One can show that in
0
new coordinates {xi } !
0
2 xl xi
 
i0
j 0 k0 = . (4.8)
xj 0 xk0 xl

Thus, only if the new coordinates are linear functions of the old Cartesian ones the new connection
coefficients will remain vanishing. Otherwise, they will not.
From eqs (4.7-4.8), it follows that the connection of Euclidean space is always symmetric with
respect to its lower indexes:
i jk = i kj (4.9)
4.2. PARALLEL TRANSPORT OF TENSORS 39

4.1.3 Riemannian Connection


Since we cannot introduce global Cartesian coordinates on Riemannian manifolds we need a different,
more general way of fixing their connections and, hence, their parallel transport. We require
the scalar product of any two vectors to remain unchanged by parallel transport;

u v = u v (4.10)

the connection to be symmetric relative to its lower indexes.


Note that both these conditions are satisfied by the parallel transport of Euclidean space.
From condition (4.10) one finds that
gij
= glj l im + gil l jm (4.11)
xm
From this result and (4.9) it follows that
 
1 gij gjm gim
jim = + , (4.12)
2 xm xi xj
where
jim = gjl l im (4.13)
and
l im = g lk kim . (4.14)
jim are called Christoffels symbols of the second kind.

4.2 Parallel transport of tensors


4.2.1 Scalars
Scalars can be considered as tensors of zero rank. The only meaningful parallel transport of scalars
is fully defined by
f = f. (4.15)

4.2.2 1-forms
Since q(u) = qi ui is a scalar it makes sense to define the parallel transport of 1-forms in such a way
that qi ui remains unchanged, that is
qi ui = qi ui . (4.16)
From this condition we obtain
qi = qi + l ij ql dxj . (4.17)

4.2.3 General tensors


Similar condition is used to define the parallel transport of tensors. For example, consider tensor
T ij . Since T ij qi uj is a scalar we require

T ij qi uj = T ij qi uj . (4.18)

This leads to
T ij = T ij i km T kj dxm + kjm T ik dxm . (4.19)
Similarly, for tensor Fij we require

Fij v i uj = Fij v i uj (4.20)


40 CHAPTER 4. GEOMETRY OF RIEMANNIAN MANIFOLDS

which leads to
Fij = Fij + kim Fkj dxm + kjm Fik dxm . (4.21)

The general rule which applies to tensors of any type can be described as follows:
The number of indexes equals to the number of terms involving ;
Each upper index is treated as a vector index;
Each lower index is treated as a 1-form index.

4.2.4 Metric tensor


According to the general rule the parallel transport of the metric tensor leads to
gij v i uj = gij v i uj . (4.22)
On the other hand, the condition eq.4.10 reads
gij v i uj = gij v i uj . (4.23)
Thus, we have
gij = gij . (4.24)

This tells us that the metric tensor gij parallel transported to the point S from the point P is
identical to the metric tensor gij already defined at S. In other words one can think of the metric
tensor as first defined at one particular point of the manifold and then parallel transported to all other
points. (This is similar to manufacturing standard metric tools in a factory and then distributing
them over the country, the planet, the Galaxy etc.)

4.3 Absolute and covariant derivatives


Definition: A tensor-valued function defined on a manifold is called a tensor field. At every point
of the manifold it defines a tensor of the same type.
On any metric manifold there defined at least one tensor field - the metric tensor field. Com-
ponents of the metric tensor in the induced coordinate basis of some local coordinates may vary
(recall gij of a sphere, Sec.1.1.2). However, as we have just discussed in Sec.3.2.4 this tensor field is
constant in the absolute sense (in the sense of parallel transport). Similarly, spherical components
of a vector field in Euclidean space vary even if this is a constant vector field. This tells us that
the usual coordinate derivatives of tensor components, like ai /xk , cannot be used to describe the
variation of tensor fields in the absolute sense (in the sense of parallel transport). For this purpose
there exist other kinds of derivatives.
4.3. ABSOLUTE AND COVARIANT DERIVATIVES 41

4.3.1 Absolute and covariant derivatives of vector fields

Consider vector field a(xk ). Parallel transport vector a from the point S to the infinitesimally
close point P (see the figure above). The result is the vector a at point P . Denote the difference
between a and a as Da:

Da = a a.
Note that Da is a vector. If Da = 0 then we say that a is the same at P and S in the absolute sense.
From eq.(4.6) it follows that
ap (xi ) = ap (xi + dxi ) + pjk (xi )aj (xi )dxk .
(Here we have sign + because the transport occurs in the direction opposite to dx.) Thus,

Dap = dap + pjk aj dxk . (4.25)


where
dap = ap (xi + dxi ) ap (xi )
as usual. If the parallel transport is carried out along the curve xp = xp () then Da/d describes
the rate of change of the vector field a(xi ) along this curve. It is called the absolute derivative of a.
One has
Dap dap dxk
= + pjk aj . (4.26)
d d d
One can rewrite eq.(4.25) as
Dap = k ap dxk . (4.27)
where
ap
k ap = + pjk aj (4.28)
xk
is called the covariant derivative of a . Since Dap and dxk are vectors defined at the same point
of the manifold then equation (4.27) is a proper tensor equation and, thus, the covariant derivative
is a second rank tensor (see Sec.2.3). This tensor describes how fast this vector field varies in all
directions (recall the gradient of a scalar function).

Note on notation:

ai
1) m ai ai ,m . (4.29)
xm
2) ai ;m m ai . (4.30)
3) ai;m m ai = g mk ai ;k g mk k ai . (4.31)
42 CHAPTER 4. GEOMETRY OF RIEMANNIAN MANIFOLDS

4.3.2 Absolute and covariant derivatives of 1-form fields


Similarly one obtains the following results for 1-forms
qp (xi ) = qp (xi + dxi ) jpk (xi )qj (xi )dxk .

Dqp = qp qp = dqp jpk qj dxk . (4.32)


where
dqp = qp (xi + dxi ) qp (xi ).
is a 1-form. The absolute derivative of q is
Note that Dq
Dqp dqp dxk
= jpk qj . (4.33)
d d d

Dqp = k qp dxk . (4.34)


where
qp
k qp = jpk qj (4.35)
xk
is the covariant derivative of q . It is a second rank tensor which describes how fast the 1-form field
varies in all directions.

Note on notation:
qi
1) m qi qi,m . (4.36)
xm
2) qi;m m qi . (4.37)
;m m mk mk
3) qi qi = g qi;k g k qi . (4.38)

4.3.3 Absolute and covariant derivatives of general tensor fields


The same procedure applies to higher rank tensors. For example, for the field of second rank tensor
T ij one obtains
DT ij dxm
= m T ij (4.39)
d d
where
T ij
m T ij = + i km T kj kjm T ik . (4.40)
xm
General rule:
The absolute derivative of a tensor field of rank r is a tensor or rank r;
The covariant derivative of a tensor field of rank r is a tensor or rank r + 1;
The first term in the expression for the covariant derivative is the usual partial coordinate
derivative (/xm ) of tensors components;
There are r more terms in this expression, one per each index. In each such term for an upper
index this index is treated as a vector index and in each such term for a lower index it is
treated as a 1-form index.
One more example
DT ijs dxm
= m T ijs (4.41)
d d
T ijs
m T ijs = + i km T kjs kjm T iks ksm T ijk . (4.42)
xm
4.4. GEODESICS AND PARALLEL TRANSPORT 43

4.3.4 Absolute and covariant derivatives of scalar fields


From eq.(4.15) it follows that for a scalar field f (scalar function)

Df dxm
= m f (4.43)
d d
f
m f = (4.44)
xm

4.3.5 General properties of covariant differentiation


For any tensors A and B of the same type

m (A + B) = m A + m B, (4.45)
and
m (AB) = (m A)B + A(m B), (4.46)
where multiplication can be both inner and outer. Although the actual number and position of
indexes of A and B does not matter here (this is why their indexes are not shown) the general rules
of tensor equations still applies.
Examples:
m (Ai j + B ij ) = m Ai j + m B ij ,

m (Ai B i ) = (m Ai )B i + Ai (m B i ),
m (Ai Bj ) = (m Ai )Bj + Ai (m Bj ),

4.3.6 The field of metric tensor


Since
gij = gij
(see Sec.3.2.4) one has
Dgij
=0 (4.47)
d
along any curve xi = xi () and
m gij = 0. (4.48)

4.4 Geodesics and parallel transport


We already know (see Sec.1.4) that geodesics are solutions of the Euler-Lagrange equations

d L L
=0 (k = 1, 2, ..., n) (4.49)
d xk xk
with Lagrangian
L(xk , xk ) = gij xi xj . (4.50)
(Recall that xk = dxk /d where is a normal parameter of the geodesic.) It easy to see that

L gij i j L
= x x and = 2gik xi .
xk xk xk
Substitution of these results into eq.(4.49) gives us

d gij i j
(2gik xi ) x x = 0.
d xk
44 CHAPTER 4. GEOMETRY OF RIEMANNIAN MANIFOLDS

gik j i gij i j
2 x x + 2gik xi x x = 0.
xj xk
 
1 gik gjk gij
gik xi + + xi xj = 0.
2 xj xi xk
Now we can use eq.(4.12) and write this result as

gik xi + kij xi xj = 0.

By raising index k (see eq.4.14 and eq.3.44) this is turned into the so-called geodesic equation

xk + kij xi xj = 0 (4.51)

which is the same as


Dti
= 0, (4.52)
d
where
dxi
ti = (4.53)
d
is the tangent vector to the geodesic. In other words, the tangent vector ti is the same along the
geodesic in the absolute sense ( in the sense of parallel transport along the geodesic). This results
allows to give the following alternative definition of a geodesic curve
Definition. A curve is called geodesic if it allows a parameter such that

Dti dxi
= 0, where ti =
d d
Such parameter is called normal and ti is called the normal tangent vector. It is this property
of geodesics that is meant when they are described as the straightest possible curves.

4.5 Geodesic coordinates and Fermi coordinates


4.5.1 Geodesic coordinates
By definition, for any point of a Riemannian manifold one can find such a system of coordinates,
called locally Cartesian that at this point

1 if i = j
gij = (4.54)
0 if i 6= j

Moreover, for any point of a Riemannian manifold one can find such a system of coordinates that

i jk = 0, (4.55)
and, hence,

gij,k = 0; (4.56)


m = ; (4.57)
xm
D d
= . (4.58)
d d
at this particular point. Such coordinates are call geodesic coordinates.
Here is how geodesic coordinates coordinates can be set up. Select a point on the manifold where
the conditions (4.55-4.58) are to be satisfied. At this point, introduce a set of basis vectors, {ei },
4.5. GEODESIC COORDINATES AND FERMI COORDINATES 45

which will become the coordinate basis of geodesic coordinates. Select a neighbourhood, Np , of P
such that for any point A Np there exists one and only one geodesic connecting it to P . Let be
such a normal parameter of this geodesic that = 0 at P . Denote as u = d/d its tangent vector
at P and as A the value of at A. Then the geodesic coordinates of point A are defined via

xiA = ui A . (4.59)

Obviously, there many normal parameters which satisfy the above selection criteria and we need to
show that the result is the same for any of them. Consider another such normal parameter, . Then

= c where c = const

and the new tangent vector


dxi 1 dxi
vi = = .
d c d
Thus,
1 i
xiA = v i A = u cA = ui A .
c
Next we need to show that in these coordinates the Christoffel symbols vanish at the point P.
From eq.(4.59) it follows that all geodesics passing through P satisfy

xi = ui where ui = const, (4.60)

which ensures
xi = 0.

Given this result the geodesic equation (4.51) reads

i jk xj xk = 0.

Thus, that for any vector ui at point P

i jk uj uk = 0

which can only be satisfied if


i jk = 0.

Geodesic coordinates are very convenient for many analytical calculations.


46 CHAPTER 4. GEOMETRY OF RIEMANNIAN MANIFOLDS

4.5.2 Fermi coordinates


In Euclidean space equations (4.55-4.58) are satisfied throughout the whole space when we employ
Cartesian coordinates (or coordinates related to the Cartesian ones via linear transformation.) For
general Riemannian manifolds it in impossible to find such coordinates that equations (4.55-4.58)
are satisfied throughout the whole manifold. The most what can be achieved in general is to get
them satisfied along a given geodesic. The corresponding coordinates are called Fermi coordinates.
Here is the way of constructing such coordinates. First we select a geodesic curve with normal
parameter such that at point O ,the origin of the Fermi coordinates, = 0 (we shell call it the
Fermi geodesic). At this point select such a basis {ei } that e1 = d/d. Parallel transport this basis
(along the Fermi geodesic) to every other point of the Fermi geodesic. Select such a neighbourhood
of the Fermi geodesic, N , that for any point A N there exists one and only one geodesic with
normal tangent vector u = d/d which connects this point to some point P of the Fermi geodesic
so that u = ui ei with u1 = 0 at P . Choose such normal parameter that = 0 at P. The Fermi
coordinates of the point A are then defined as

x1 = P
(4.61)
xi = A ui i = 2, . . . , n.

In these coordinates the Fermi geodesic satisfies the equation

d2 xi
=0
d2
and the geodesic through A satisfies
d2 xi
= 0.
d2
This ensures that for any geodesic through P

i jk xj xk = 0

which can only be satisfied if


i jk = 0.
Fermi coordinates play important role in the theory of relativity. They correspond to the so-called
free-falling frames.
4.6. RIEMANN CURVATURE TENSOR 47

4.6 Riemann curvature tensor


Parallel transport on Riemannian manifolds has a number of properties not seen in Euclidean space.
This is clearly demonstrated in the following examples involving a 2D sphere. Recall that any vector
tangent to a geodesic remains tangent during parallel transport along this geodesic. Moreover, since
the angle between two parallel transported vectors is constant so must be the angle between a vector
parallel transported along a geodesic and this geodesic.

1. The result of parallel transport depends not only on the initial and final points but also on the
path along which this transport is carried out!

(a) When vector t is parallel transported from the point A on the equator to the north pole,
N , along the meridian AN the result is vector t0 ;
(b) When vector t is first parallel transported from the point A to the point C along the
equator, which results in vector t , and then parallel transported from C to N along the
meridian CN the result is a different vector, t00 6= t0 .

2. Parallel transport along a closed curve does not result in the original vector!
Indeed, when vector t0 is parallel transported along the closed path N ACN the result is vector
t00

Obviously these peculiar properties stem from the fact that sphere is a curved surface! Curvature
of such surfaces and general manifolds is described via the so called Riemann curvature tensor.
48 CHAPTER 4. GEOMETRY OF RIEMANNIAN MANIFOLDS

Consider a manifold M and a point A M. Select vectors a, dx(1) , and dx(2) defined at A.
Introduce local coordinates {xi } and construct the close path ABCDA as shown in the figure.

Parallel transport vector a along this path (first in the direction of dx(1) ) to obtain vector a + da
at point A. Since this path is infinitesimally small da~ must depend linearly on ~a, dx ~ (1) , and dx
~ (2)
~ ~
and vanish if ~a = 0 or dx(1) = 0 or dx(2) = 0. That is we must have

p
dai = Ri lmp al dxm
(1) dx(2) . (4.62)

Since this is a proper tensor equation, Ri lmp is a tensor and it is called the Riemann curvature
tensor. Direct calculations show that

Rklmp = klm,p + klp,m slm kps + slp kms (4.63)

(Note that although i jk is not a tensor, Rkijl is(!) Such peculiar results do occur from time to
time.)

Curvature of manifolds also causes deviation of initially parallel geodesics. Consider two
infinitesimally close points, A and B, separated by the infinitesimal displacement vector dx.
4.7. PROPERTIES OF THE RIEMANN CURVATURE TENSOR 49

~ Construct two geodesics


Select some vector t at A and parallel transport it from A to B along dx.
passing through A and B with normal parameter . Namely, the geodesic xi = xi(A) () such that

dxi
=0 and = ti at A
d

and the geodesic xi = xi(B) () such that

dxi
=0 and = ti at B .
d
These geodesics can be described as parallel at points A and B. Denote the displacement vector
separating the points of these two geodesics which have the same value of as

sd where d = const.

One can show that


Dsi dxl dxj k
= Ri ljk tl tj sk Ri ljk s . (4.64)
d d d
This equation in called the equation of geodesic deviation. It shows that initially parallel geodesics
deviate from each other.

In Cartesian coordinates of Euclidean space all i jk = 0 and from (4.63) one has

Ri jkl = 0.

Since R is a tensor, this is true in any basis ( R is just a zero tensor.) Thus, all dai in (4.62) and all
Dsi /d in (4.64) vanish and we recover the familiar properties of Euclidean space.

Definition A manifold is called internally flat (often just flat) if everywhere on this manifold
Ri jkl = 0, otherwise it is called internally curved.

For example planes and cylinders of Euclidean space are internally flat manifolds (surfaces).

One can also show that


(m p p m )ak = Rklmp al ; (4.65)
l
(m p p m )ak = R kmp al . (4.66)
Thus, on curved manifolds the operators of covariant differentiation do not commute.

4.7 Properties of the Riemann curvature tensor


Ri jkl has a number of properties which reduce the number of its independent components:

Rpijk = Rpikj ; (4.67)

Rpijk = Ripjk ; (4.68)


Rpijk = Rjkpi ; (4.69)
Rppij = 0; (4.70)
Rpijk + Rpjki + Rpkij = 0. (4.71)
50 CHAPTER 4. GEOMETRY OF RIEMANNIAN MANIFOLDS

Note the cyclic permutation of the lower indexes in eq.(4.71). The best way of proving these
properties involves use of geodesic coordinates. Indeed, since in geodesic coordinates i jk = 0
and gij,k = 0, eq.(4.63) has a much simpler form

Rklmp = klm,p + klp,m . (4.72)

Using eq.(4.12) this can also be written as


1
Rklmp = [gkp,lm + glm,kp gkm,lp glp,km ] (4.73)
2
(By the way, eq.(4.73) tells us that all second order derivatives of gij vanish only if Ri jkl = 0!) Now
it easy to see that, for example,

Rklmp = klm,p + klp,m = (klm,p klp,m ) = (klp,m + klm,p ) = Rklpm ,

which proves eq.(4.67).


Because of the properties (4.67-4.71) the curvature tensor has only
1 2 2
N= n (n 1)
12
independent components (the total number of components is n4 , where n is the dimension of the
manifold.) If n = 2 then N = 1. The curvature tensor of 2D manifolds has only one independent
component.
One can also show (using geodesic coordinates once again) that

Ri kpl;m + Ri klm;p + Ri kmp;l = 0 (4.74)

(Note the cyclic permutation of indexes p, l, m in this equation.) This result is known as the Bianchi
identity.

4.8 Ricci tensor, curvature scalar and the Einstein tensor


These important tensors are derived from the Riemann curvature tensor. The Ricci tensor is defined
via
Rij = Rsisj (4.75)
The symmetries (4.67-4.69) ensue that

Rij = Ri ssj = Rsijs = Rsisj , (4.76)

as well as
Rij = Rji . (4.77)
The curvature scalar is defined as
R = Ri i . (4.78)
The Einstein tensor is
1
Gij = Rij Rgij . (4.79)
2
It is easy to see
Gij = Gji . (4.80)
Moreover, using the Bianchi identity one can show that

Gik;k k Gsk 0. (4.81)

In other words, the divergence of Einsteins tensor is zero.


Chapter 5

Space and time in the theory of


relativity

Each physical theory is based on a number of key assumptions - the rest of the theory is then build
on these assumptions using appropriate mathematical tools. Any good theory has to be

1. self-consistent (no internal contradictions),

2. consistent with Nature.

It seems like the second condition can never be achieved completely. As we learn more we discover
new, previously unknown contradictions between our theories and Nature. They force us to revise
our theories by constructing new sets of basic assumptions.
What was revised during the transition from from Newtonian physics to General Relativity are
the assumptions on the nature of physical time and space.

5.1 Physical Space and Time in Newtonian Physics


Time: Physical time is absolute it is the same everywhere in the Universe. Thus, it can be
measured by a single standard clock and can be modeled as a one-dimensional metric space
with coordinate t. This t is a convenient universal parameter for trajectories of particles.

Space: Physical space is also absolute - it is the same at any time. For any moment of time
one can introduce Cartesian coordinates covering the whole physical space. Thus, this is a
Euclidean space.

Inertial frames: In order to describe motion/evolution of physical systems we need to have


a system of spatial coordinates at any time. This can be arranged by building a spatial grid
- a system of physical objects (nodes) which may or may not be kept at the same distance
from each other. Such spatial grid is called a reference frame. Depending of the problem
under consideration some of these references frames are better than others as they provide a
simple description of the studied phenomena. In particular, the motion of free particles has
a particularly simple form when described in so-called inertial frames where they move with
constant speed:

Dv i
= 0, (5.1)
dt

51
52 CHAPTER 5. SPACE AND TIME IN THE THEORY OF RELATIVITY

where
dxi d
vi = or v = .
dt dt
One can think of an inertial frame as a collection of free particles moving with the same speed,
so that the distances between them are fixed, the grid of spatial coordinates being attached to
these particle. These spatial coordinates do not have to be Cartesian the use of the absolute
derivative in eq.5.1 ensures that the choice of spatial coordinates is not important. Since
eq.5.1 in nothing else but the geodesic equation, the trajectory of free particle is a geodesic of
Euclidean space. The absolute time t is a normal parameter of this geodesic, and v i = dxi /dt
is its tangent vector.
In Cartesian coordinates eq.5.1 can be written as

dv i d2 xi
= 0, or = 0.
dt dt2

If the particle is subject to a force f the law of motion becomes

D(mv i )
= f i. (5.2)
dt
where scalar m is the particles mass.

Newtonian principle of relativity:


There are infinitely many inertial frames and there is no way to tell which of them are moving
and which of them are at rest in space. All laws of mechanics are exactly the same for all of
them. Thus, there is no absolute motion, that is a motion relative to the space itself, but only
the relative motion, that is a motion of some objects relative to others.

5.2 Physical Space and Time in Special Relativity


Time, Space and Inertial Frames. Various experiments show that in vacuum electromag-
netic waves propagate with the same speed, c, relative to all inertial observers. Moreover,
c is the maximum possible speed for any signal. This implies that time can no longer be
considered as absolute because there is no unambiguous way of determining simultaneity of
spatially separated events. (Only if there existed a way of communicating information with
infinitely large speed then simultaneity would have an absolute meaning.) This also explains
why moving relative to each other inertial observers measure different time intervals between
the same events even if they use identical clocks and identical procedures. Because, time is
no longer absolute, an inertial frame of Special Relativity is no longer just a collection of free
particles moving with the same velocity and carrying the spacial grid. They should also carry
synchronized clocks used to measure time of this particular frame.
Physical space is not absolute either as different inertial observers obtain different results
when they measure lengths of the same objects even if they use identical standard meters and
identical procedures. This implies that physical space can not be described as a metric space.
However, all inertial observers obtain the same result for the spacetime interval

s2 = c2 t2 + l2 (5.3)
for any two events. This suggests to unite physical time and space into a single 4-dimensional
metric space, called spacetime, and consider s as a generalised distance between its points,
called events. If s2 > 0 then there exists a frame where t = 0 and s2 = l2 . Such
spacetime intervals are called space-like. If s2 < 0 then there exists a frame where l = 0
and s2 = c2 t2 . Such spacetime intervals are called time-like. Spacetime intervals such
5.2. PHYSICAL SPACE AND TIME IN SPECIAL RELATIVITY 53

that s2 = 0 are called null. They cannot be reduced to either pure space or pure time
intervals. (If fact they describe events on the world-line of a light signal.)
Here are the basic assumptions on the nature of spacetime in Special Relativity:
1. Spacetime is a 4-dimensional metric space. If {x }, = 0, 1, 2, 3, are arbitrary coordinates
of the spacetime then the interval between its infinitesimally close points is given by the
metric form

ds2 = g dx dx where , = 0, 1, 2, 3. (5.4)


2. There exist such coordinate systems that
(a) The hyper-surfaces x0 =const have purely space-like metric form

dl2 = gij dxi dxj i, j = 1, 2, 3, (5.5)


0
with gij independent on x ;
(b) The spacetime metric (5.4) reduces to

ds2 = (dx0 )2 + dl2 . (5.6)


Such coordinate systems correspond to inertial frames with spatial coordinates xi ,
i = 1, 2, 3, and global time
t = x0 /c.
(c) There exist such spatial coordinates xi that that

dl2 = (dx1 )2 + (dx2 )2 + (dx3 )2

throughout the whole hypesurface x0 =const. Thus, this hypersurface is a 3-dimensional


Euclidean space.
In other words, spacetime allows coordinates {x } such that

ds2 = (dx0 )2 + (dx1 )2 + (dx2 )2 + (dx3 )2 (5.7)

and, hence,

1 0 0 0
0 1 0 0
g =
0
(5.8)
0 1 0
0 0 0 1
throughout the whole spacetime. These coordinates are called pseudo-Cartesian. Metric
spaces which allow such coordinates are called pseudo-Euclidean (or Minkowskian).
Einsteins principle of relativity:
Einstein assumed that not only mechanical laws but all physical laws are exactly the same in
all inertial frames. Thus, the idea of absolute motion is completely rejected. This is known as
Einsteins principle of relativity. Introduction of spacetime allows us to transform this principle
into a simple prescription for writing relativistic laws of physics. Indeed, as we have already
seen, different inertial frames correspond to different systems of coordinates in spacetime. Thus
if we write laws of physics in the form which does not involve coordinates whatsoever then
they will automatically satisfy the principle of relativity. Tensor equations involving spacetime
tensors fit into this category perfectly well.
For example, consider the motion of free particles. In spacetime each particle traces a curve
which is called its world-line. One can use the particles proper time, , as a parameter of the
curve. The tangent vector
d dx
~u = or u = (5.9)
d d
54 CHAPTER 5. SPACE AND TIME IN THE THEORY OF RELATIVITY

is called the 4-velocity of the particle. Note that this is a spacetime vector. Now consider the
following tensor equation

Du
= 0. (5.10)
d
In the pseudo-Cartesian coordinates of any inertial frame this reads
du
=0 or u = const. (5.11)
d
Thus,
dt dxi
c = const and = const,
d d
which gives us
dxi dv i
= const or = 0. (5.12)
dt dt
Since in Cartesian coordinates
dv i Dv i
=
dt dt
we have recovered the 3-tensor equation (5.12) of motion of free particles:

Dv i
= 0.
dt

Thus, (i) the 4-tensor eq.(5.10) is the law of motion of free particles; (ii) from this 4-tensor
equation there follows the 3-tensor law of motion which has the same invariant form for all
inertial frames (in agreement with the principle of relativity).

5.3 Relativistic equations of motion of particle dynamics


The 4-tensor equation of motion of free particles (eq.5.10),
Du
= 0,
d
is nothing else but the geodesic equation. Thus, world lines of free particles are geodesics of
spacetime. Since u u < 0 they are time-like geodesics. Using the spacetime connection this
equation can be written as
du
+ u u = 0. (5.13)
d
and, finally, as a system of second order PDEs for x ( )

d2 x dx dx
2
+ = 0. (5.14)
d d d
(This is the most general form of equations of motions of free particles which holds for any
system of spacetime coordinates.)
In pseudo-Cartesian coordinates all Christoffels symbols vanish and eq.(5.14) reduces to

d2 x
= 0, (5.15)
d 2
which immediately integrates to give

x = a + b where a , b = const.
5.4. CONSERVATION LAWS 55

This has the same form as the equation of straight lines of Euclidean space in Cartesian
coordinates.
When a particle is subjected to a force its 3-velocity is no longer constant and neither is its
4-velocity. The appropriate modification of (5.10) is
D(mu )
= f, (5.16)
d
where f is a spacetime vector called the four-force and m is the mass of the particle as
measured in the frame where it is at rest. Hence the name, the rest mass. This definition
ensures m is the same for all inertial frames and, hence, that m is a spacetime scalar.
If the case of the electromagnetic force

q
f =F u , (5.17)
c
where q is the electric charge of the particle (a spacetime scalar), and F is the electromagnetic
field tensor.

5.4 Conservation laws


Consider a continuous medium that can be attributed with some scalar quantity M of volume density
. The amount of M within volume V is then
Z
M = dV
V

~ gives us the flux of M across the surface element


Vector J~ is called the flux density of M if J~ dS
~ in the direction shown by dS,
dS ~ that is the amount of M passing through the surface element per
unit time. The total amount of M leaving volume V is then simply
Z Z
J dS = J i dSi ,
V V
~ is its outgoing surface element.
where V is the surface of V and dS

dS

 






 

 

 

 

 

 

 






 

 
 

 
 

 
 

 
 

 
 

 
 




 
      
 J

Figure 5.1: Integration volume V , its surface element dS, and the flux density vector J.

If M is not created or destroyed inside V then the amount of M in this volume varies only due
to the flow of M out of V into the outside space. Hence, we have
Z Z
d
dV = J dS
dt
V V

or Z Z
d
dV + J dS = 0. (5.18)
dt
V V
56 CHAPTER 5. SPACE AND TIME IN THE THEORY OF RELATIVITY

This is the integral form of the conservation law for the scalar quantity M with volume density
~ According to the Gauss theorem one can rewrite this as
and flux density J.
Z Z
d
dV + i J i dV = 0
dt
V V

and, thus, Z  
i
+ i J dV = 0.
t
V
Since the volume V is arbitrary we deduce from this that

+ i J i = 0, (5.19)
t
which is called the differential form of the conservation law for scalar quantity M with volume
~
density and flux density J.
If replace M with a vector quantity then in the place of we should have a vector, e.g. P , and
in the place of J~ we should have a tensor, e.g. T ij . The integral conservation law will then look as
Z Z
d
P i dV + T ij dSj = 0. (5.20)
dt
V V

and the differential one as


P i
+ j T ij = 0. (5.21)
t
Notice, that eq.5.21 is a proper tensor equation whereas eq.5.20 is not because it involves addition
of components of vectors defined at different points of space.
Sometimes it is relatively easy to figure out the flux density. For example, consider a swarm of
particles of mass m, number density n and 3-velocity ~v as measured in some inertial frame. Suppose
that the number of particles is conserved. Then the total mass of the swarm will also be conserved.
~ are these that occupy
During the time interval (t, t + dt) the only particles that cross the surface dS
at time t the oblique cylinder shown below.
 

   
vdt 
  
 
dl= 
v

  

   
  
 
  dS

   
  

Figure 5.2:

Its volume is
dV = v dt dS = (v dS)dt;
the total number of particles in this volume is dN = ndV and the total mass is dM = mndV . Thus,
the total mass carried through the surface dS during the time interval dt is
dM = nm(v dS)dt.
This shows that the mass flux density is
J = nmv. (5.22)
5.5. RELATIVISTIC CONTINUITY EQUATION 57

5.5 Relativistic continuity equation


Consider a swarm of particles moving with four-velocity u . Let n be the number density of these
particles as measured in the inertial frame moving with the same velocity as the local velocity of the
swarm. It is called the proper number density. Consider the following 4-tensor equation

(nu ) = 0. (5.23)

Here is the operator of covariant differentiation in spacetime. Notice that n is a spacetime scalar,
(nu ) is a spacetime tensor, and hence (nu ) is a spacetime scalar. Thus, equation (5.23) is
a proper tensor equation and may express a physical law (see the principle of relativity). But what
law? In the pseudo-Cartesian coordinates of an arbitrary inertial frame, laboratory frame,

= and u = (c, v 1 , v 2 , v 3 ) (5.24)
x
where is the Lorentz factor and v i is the usual velocity vector (three-vector). Thus, (5.23) reads

(cn) + (nv i ) = 0 (5.25)
x0 xi
Since x0 = ct and in Cartesian coordinates

i =
xi
this can also be written as

(n) + i (nv i ) = 0 (5.26)
t
or

(n) + i (nv i ) = 0 (5.27)
t
where n = n is the number density of particles as measured in the laboratory frame, it is different
from n because of the Lorentz contraction. (Notice that i is the operator of covariant differentiation
in space, the hypersurface x0 =const.) Obviously, eq.(5.26) describes the conservation of particles as
seen in the laboratory frame and, thus, the 4-tensor equation (5.23) describes the same conservation
but in a coordinate independent form.
Introduce the proper rest mass density of the swarm,

= mn. (5.28)

Since the rest mass m is a spacetime scalar we can now rewrite (5.23) as

(u ) = 0. (5.29)

This 4-tensor equation is called the relativistic continuity equation.

5.6 Stress-energy-momentum tensor


5.6.1 Energy-momentum vector
Consider a particle of rest mass m and 4-velocity u . The 4-vector

P = mu (5.30)

is called the energy-momentum vector of the particle (or its4-momentum vector). In the laboratory
frame
P = m(c, v 1 , v 2 , v 3 ) = (E/c, p1 , p2 , p3 ), (5.31)
58 CHAPTER 5. SPACE AND TIME IN THE THEORY OF RELATIVITY

where
E = mc2 (5.32)
is the particle energy and
pi = mv i (5.33)
is the 3-vector of particles momentum as measured in the laboratory frame.

m = m (5.34)

is called the inertial mass of the particle, its mass as measured in the laboratory frame. From
eq.(5.12) it follows that for a free particle

E = const and pi = const. (5.35)

Thus, the energy and the momentum of a free particle are conserved. The 4-tensor equation with
describes this conservation is

DP
= 0. (5.36)
d

5.6.2 Stress-energy-momentum tensor of dust


Consider now a swarm of free particles of the proper number density n, the proper rest mass density
= mn, and the 4-velocity u . The tensor

T = u u (5.37)

is called the stress-energy-momentum tensor of the swarm. Components of this tensor also allow
simple interpretation. Consider an arbitrary inertial frame (the laboratory frame). Denote as e the
energy density (per unit volume), as i the momentum density and as si the energy flux density of
the swam in this frame. Then

e = T 00 , i = (1/c)T 0i = (1/c)T i0 , si = cT 0i = cT i0 . (5.38)


Moreover, T ij if the momentum flux density, the stress 3-tensor. (T ij gives us the flux density of
the i-component of momentum in the j-direction, and, at the same time, the flux density of the
j-component of momentum in the i-direction.) Indeed,


e = u0 u0 = 2 c2 = (mc2 )(n) = E n; (5.39)


i = u0 ui /c = 2 v i = (mv i )(n) = pi n; (5.40)


si = cu0 ui = 2 c2 v i = ev i ; (5.41)


T ij = ui uj = ( 2 v i )v j = i v j ; (5.42)
and
T ij = ui uj = ( 2 v j )v i = j v i . (5.43)

Thus, like the usual energy and momentum of a single particle are simply components of a first
rank 4-tensor ( energy-momentum vector), the usual energy density, momentum density, energy flux
density and stress tensor (3-tensor) of a continuously distributed system are components of a second
rank 4-tensor (stress-energy-momentum tensor).
5.6. STRESS-ENERGY-MOMENTUM TENSOR 59

5.6.3 Energy-momentum conservation


Consider the following equation

T = 0. (5.44)
Since this is a 4-tensor equation it may express some physical law. But what law? In the pseudo-
Cartesian coordinates of the laboratory frame

= . (5.45)
x
Thus, eq.(5.44) reads

T =0 (5.46)
x
or
0 i
0
T + T =0 (5.47)
x xi
or
1 0 i
T + T = 0. (5.48)
c t xi
The time component of this equation ( = 0) can be written as

e + i si = 0. (5.49)
t
This is just the energy conservation law. The spatial component ( = 1, 2, 3) of eq.(5.48) reads as
j
+ i T ji = 0. (5.50)
t
This is just the momentum conservation law. Thus eq.(5.44) describes the conservation of energy
and momentum in a coordinate independent form.

5.6.4 Stress-energy-momentum tensor of perfect fluid


Any continuous system like a fluid or a force field can be attributed with its own stress-energy-
momentum tensor and if this system is isolated (does not interact with other systems) then eq.(5.44)
is satisfied.
Let us determine the stress-energy-momentum tensor of ideal fluid. What we need is an expres-
sion for this 4-tensor in terms of lower rank (more basic) 4-tensors like in eq.(5.37). But it is easier
to figure out the components of T ij in the rest frame of the fluid. (By this we mean the inertial
frame where the fluid is at rest. Obviously, each fluid element has it own rest frame.) In the rest
frame the energy per unit volume is given by

T 00 = e = c2 + ,

their is the rest mass density and  is the thermal energy density. Moreover, since v i = 0 the
momentum density and hence the energy flux density vanish in this frame

i = T 0i = T i0 = 0.

The components of stress tensor of ideal fluid at rest in Cartesian coordinates are

p 0 0
T ij = 0 p 0
0 0 p

where p is the thermodynamic pressure. Thus, in the pseudo-Cartesian coordinates of the rest frame
the components of stress-energy momentum tensor of ideal fluid are
60 CHAPTER 5. SPACE AND TIME IN THE THEORY OF RELATIVITY


e 0 0 0
0 p 0 0
T =
0
. (5.51)
0 p 0
0 0 0 p
Note that e and p are spacetime scalars. They completely determine the thermodynamical state
of ideal fluid. Its motion is completely determined by the 4-vector u . Thus, what we need to do
now is to construct T from e, p, and u via suitable tensor operations in such a way that in the
fluid frame we end up with eq.(5.51). In fact,
e+p
T = ( )u u + pg (5.52)
c2
does the job. Indeed, in the pseudo-Cartesian coordinates of the rest frame of the fluid

u = (c, 0, 0, 0)

1 0 0 0
0 1 0 0
g =
0 0 1
,
0
0 0 0 1
and eq.(5.52) reduces to eq.(5.51).
5.7. SPACE AND TIME IN GENERAL RELATIVITY 61

5.7 Space and Time in General Relativity


The key idea of General Relativity is that gravitational interaction makes itself felt via producing
internal curvature of spacetime. The following example from Newtonian mechanics helps to under-
stand this idea. Consider the motion of particles bound to a spherical surface but otherwise free of
any force. Such particle simply move along geodesics of the sphere (with constant speed). These
geodesics are great circles. Consider two such particles initially located on the equator with parallel
and equal initial velocities. As they move in the direction of the north pole they accelerate toward
each other as if they were under the action of a mutual attraction force.

Space and Time: The spacetime is no longer a flat pseudo-Euclidean space but a curved
pseudo-Riemannian manifold. This manifold is only locally pseudo-Euclidean which means
that one can introduce local coordinates such that

1 0 0 0
0 1 0 0
g =
, (5.53)
0 0 1 0
0 0 0 1
at a single point but not throughout the manifold. All basic results from the theory of Rie-
mannian manifolds apply to pseudo-Riemannian manifolds.

Inertial frames: There are no global inertial frames in General Relativity. Indeed, the cur-
vature of spacetime (inflicted by gravitational interactions) does not allow us to introduce
global pseudo-Cartesian coordinates. However, one can introduce local pseudo-Cartesian co-
ordinates, like geodesic coordinates or Fermi coordinates. These correspond to locally inertial
frames. In particularly, the Fermi coordinates built around a time-like geodesic correspond
to a small free-falling laboratory. Within the small volume of such a laboratory the effects of
finite curvature of spacetime are also very small.

General Principle of Relativity: It is not possible to detect the effects of gravitational


interaction via local measurements. In other words, via experiments carried out within a
sufficiently small free-falling laboratory, one can not detect the presence of nearby gravitating
bodies. For example, the astronauts on board of a space station orbiting the Earth and the
astronauts on board of a spacecraft free flying far away from Earth, in almost empty deep
space, share exactly the same experiences, e.g. weightlessness. This implies that, all local
physical laws (laws formulated in terms of quantities defined at a single point of spacetime)
must have exactly the same form as in the flat spacetime of Special Relativity and, hence, the
Riemann curvature tensor can not appear in them. (The only exception are the equations
62 CHAPTER 5. SPACE AND TIME IN THE THEORY OF RELATIVITY

of gravitational field which show how exactly the curvature is imposed on spacetime.) For
example, the motion of free particles is still described by equation (5.1),
Du
= 0,
d
and, thus, their world lines are still geodesics of spacetime. Similarly, the motion of ideal fluid
(and dust) is still described equations (5.29,5.44),

u = 0,

T = 0.

5.8 Einsteins equations of gravitational field


Let us speculate on how the equations of gravitational field could look like. Obviously they must be
tensor equations and should involve the Riemann curvature tensor. If this curvature is caused by
matter then some tensor fields describing the distribution of matter must also be involved.
In Newtonian gravity matter is present in the form of its volume mass density. But in relativistic
physics things are different in two respects:
1. Mass is attributed not only to matter but also to force fields (like the electromagnetic field)
via m = E/c2 . This suggests that not only matter but also force field can curve spacetime.
2. The volume mass density (or energy density) is not a spacetime scalar but just one component
of the stress-energy-momentum tensor. This suggests to seek a simple tensor equation relating
the Riemann curvature tensor with T , the total stress-energy-momentum tensor!
The metric tensor may also be present in this equation because of it fundamental role in geometry.
However, the equation must agree with the symmetries of involved tensors. It does not seem possible
to relate T and g with R directly. For example, the equation

R = ag T

is in conflict with the symmetries of the Riemann curvature tensor (eqs.4.67-4.71). However, the
Ricci tensor has the same rank and the same symmetry as T and initially Einstein suggested that

R = aT ,

where a is a constant. However, he quickly realised that this is no good. Indeed, because

R 6= 0

one ends up with


T 6= 0
which contradicts to the general principle of relativity. So, Einstein suggested another equation
which is free from such a flaw, namely
G = aT , (5.54)
where G = R (R/2)g is now known as the Einstein tensor. Indeed, as we have already seen
(equation 4.81),
G 0
implying
T = 0. (5.55)
Equation (5.54) is known as the Einstein equation of gravitational field. Later we shell see that
8G
a= (5.56)
c4
5.8. EINSTEINS EQUATIONS OF GRAVITATIONAL FIELD 63

The Einstein equation (5.54) is the key equation of General Relativity. Once this equation is
introduced we can forget all the reasons which have led Einstein to this equation and simply derive
from it all the important results of the Theory of Relativity. For example, as we have already seen,
equation (5.55) follows directly from the Einstein equation. This equation describes the dynamics
of continuous media like fluids and fields. For a swarm of dust particles

T = u u , (5.57)
where = mn is the rest mass density of the swarm (see Sec.4.5.2). Hence, eq.(5.55) reads

T = u u = u u + u u = 0. (5.58)
Since these particle do not interact with each other their total number is conserved and we have

u = 0. (5.59)
This allows us to write eq.(5.58) as
u u = 0,
or
Du
= 0.
d
This is the equation of motion of free particles (geodesic motion).
Now we may consider a time-like geodesic of a free falling laboratory and construct the corre-
sponding system of Fermi coordinates of this geodesic. Since in these coordinates

D d
= 0 = and =
dt dt x
along the geodesic (notice that here = t) the equations of continuous dynamics, (5.55) and (5.59)
reduce to

T = 0,
x
and

u = 0,
x
and the equation of geodesic motion reduces to

du
= 0.
dt
This is exactly how they read in the pseudo-Cartesian coordinates of flat spacetime. Thus, gravity
disappears in free falling locally inertial frames.
Einsteins equation can be written in slightly different form which we shell use later on. To
obtain this, we first contract (eq.5.54)

1
G = aT or R R = aT
2
Next we denote
T as T (5.60)
and use that = 4 to obtain
R = aT. (5.61)
Finally, substitute this into (5.54) to obtain

1
R = a(T T g ). (5.62)
2
64 CHAPTER 5. SPACE AND TIME IN THE THEORY OF RELATIVITY

We know that components of the Riemann curvature tensor in the coordinate basis are functions
of and , . We also know that are functions of g and g, . Thus, the components of
R , and hence the components of R and G , depend on the components of g and their first
and second partial derivatives. Thus, the Einstein equations can be viewed as second order partial
differential equations for the components of the metric tensor! The total number of independent
equations in this system is 10 ( Do you know why?) The same is the total number of independent
components of the metric tensor. What a match! However, rather complicated analysis of the
Einstein equations shows that they include only 6 evolution equations that describe the time-
evolution of g . Others may be consider as differential constrains on the initial solution (like
B ~ = 0 in electrodynamics). Thus, there in no match after all and the system appears to be under-
determined. In fact, this is good news! Indeed, the components of metric tensor depend not only on
the structure of the spacetime but also on the system of coordinates we choose. When we introduce
four coordinates in spacetime we effectively impose four additional conditions on the components of
metric tensor. And in reverse, an introduction of four additional conditions on the components of
metric tensor amounts to setting up a coordinate system. Here is two examples of such conditions:
The conditions
g00 = 1, gi0 = 0.
define the so-called time-orthogonal coordinates (they may not exist).
The conditions
g = 0, which ensure x = 0,
introduce the so-called harmonic coordinates.
Often one cannot give a clear physical interpretation of coordinates introduced in such a way (e.g.
one cannot tell which coordinate is time-like and which are space-like). Only after the Einstein
equations are solved and the functions g (x ) are found such an interpretation becomes possible.
In fact, the Einstein equations are local and do not tell anything about the spacetime topology.
We have to make explicit assumptions on the topology of spacetime for example, we may assume
that it has the same topology as a 4-dimensional sphere of a 5-dimensional Euclidean space. But
will this be a correct assumption?
5.9. NEWTONIAN LIMIT 65

5.9 Newtonian limit


The Newtonian theory of gravity has been extremely successful. It describes the motion of planets
and satellites with great accuracy. This means that in the limit of low velocities and hence weak
gravity any good theory of gravity must reduce to the Newtonian theory. Let us check that the
Einstein theory satisfies this condition.
Here is the basic equations of the Newtonian theory.

The equation of motion (the second law of particle mechanics):

~ Dv i
~a = or = i , (5.63)
dt
where ~v is the particle velocity, ~a is the particle acceleration, and is the gravitational poten-
tial.

The equation of gravitational field:


= 4G, (5.64)
where is the mass density, G is the gravitational constant, and

= i i = g ij i j

is the Laplace operator.

In Cartesian coordinates {xi } these equations read

dv i
= i (5.65)
dt x
and
3
X 2
2 = 4G. (5.66)
i=1 xi
The basic equations of the Einstein theory are

The equation of motion:


Du
= 0, (5.67)
d
The field equation:
1
R = a(T T g ). (5.68)
2
What are the conditions of weak gravity?

1. The curvature of spacetime must be very small. Thus, there must be possible to construct
such a system of coordinates that the metric tensor has almost the same components as in flat
spacetime (Minkowskian) in pseudo-Cartesian coordinates. That is

g = + h , (5.69)

where
1 0 0 0
0 1 0 0
=
0
, (5.70)
0 1 0
0 0 0 1
and
|h |  1. (5.71)
66 CHAPTER 5. SPACE AND TIME IN THE THEORY OF RELATIVITY

2. Moreover, the particle velocity must be much lower than the speed of light. Hence, we may
assume
u0 = c, and |ui |  c. (5.72)
This ensures that the proper time of the particle, , is very close to the coordinate time
t = x0 /c:
= t. (5.73)
Moreover, given such low characteristic speeds
v
0
i
 . (5.74)
x c x xi

3. Finally, for a nonrelativistic thermal motion (low temperatures)

c2  , p. (5.75)

This means that only the rest mass of gravitating objects makes any noticeable contribution
to their stress-energy-momentum tensors. Thus, the gravitational field is fully determined by
the distribution of rest mass.

Conditions (5.72) and (5.75) show that the T00 component of the T tensor is much larger than
all other components and we may assume that
2
c 0 0 0
0 0 0 0
T = 0
(5.76)
0 0 0
0 0 0 0
with great accuracy.
Let us check if under these conditions the Einstein equations (5.67) and (5.68) reduce to the
Newtonian equations (5.65) and (5.66). Let us start with the equation of motion. Using (5.73) one
can write the spatial part of (5.67) as

dui
+ i u u = 0.
dt
Using (5.72) this can be written as
dui
+ i 00 c2 = 0. (5.77)
dt
From (4.12) we have  
1 g g g
= g + .
2 x x x
Substituting g from (5.69-5.71) and keeping only the terms first order in h we obtain
 
1 h h h
= + , (5.78)
2 x x x

which gives us  
1 i h0 h0 h00
i 00 = + .
2 x0 x0 x
Since is diagonal we have
   
i 1 ii hi0 hi0 h00 1 hi0 hi0 h00
00 = + = + .
2 x0 x0 xi 2 x0 x0 xi

Finally, using (5.74) we may ignore the derivative with respect to x0 and obtain
5.9. NEWTONIAN LIMIT 67

1 h00
i 00 = . (5.79)
2 xi
Then eq.(5.77) reads

dui c2 h00
= . (5.80)
dt 2 xi
Notice that this equation has exactly the same form as (5.65). This suggests to relate h00 with
Newtonian gravitational potential via

c2
= h00 . (5.81)
2
Let us now deal with the field equation. From eq.(5.76) we find

T00 = c2 and T = T = T 00 = 00 T00 = c2 .


Thus, the time component of (5.68) reads
a 2
c .
R00 = (5.82)
2
Equation (5.78) shows that all components of are small (of order O(h)). Keeping only terms
linear in we can write

R =
x x
(see eq.4.63) and thus

R00 = R00 = .
x 00 x0 0
Once again we may ignore derivatives with respect to x0 and obtain
i
R00 = .
xi 00
Substitution of i 00 from (5.79) into this equation gives us
3
1 X 2 h00
R00 = .
2 i=1 xi 2

Thus, equation (5.82) reads


3
1 X 2 h00 a
= c2 . (5.83)
2 i=1 xi 2 2

Using eq.(5.81) we can write this as


3
X 2 ac4
= . (5.84)
i2
x 2
i=1

This equation has exactly the same form as the Newtonian field equation (5.66). Thus, Einsteins
equations do reduce to the Newtonian equations indeed! Moreover, now we can express constant a
of the Einstein equations in terms the gravitational constant, G, and the speed of light:
8G
a= . (5.85)
c4
68 CHAPTER 5. SPACE AND TIME IN THE THEORY OF RELATIVITY
Chapter 6

Schwarzschild Solution

In this chapter we study a particular solution of Einsteins equations that describes the spacetime
outside of a spherically symmetric non-rotating body of a certain mass, e.g. a non-rotating black
hole, and the motion of test particles in such spacetime. Throughout this chapter we use the
relativistic units, also known as the geometric units, where G = 1 and c = 1.

6.1 Schwarzschild Solution


6.1.1 Schwarzschild Solution in Schwarzschild coordinates
The interval of pseudo-Euclidean spacetime of special relativity in pseudo-Cartesian coordinates is
given by
ds2 = dt2 + dx2 + dy 2 + dz 2
For problems with spherical symmetry in space (t =const hypersurface) it is more convenient to use
spherical spatial coordinates {r, , }. Then the spacetime interval takes the following form:

ds2 = dt2 + dr2 + r2 (d2 + sin2 d2 ).

Now, let us try to come up with a simple and reasonable expression for the metric form of the
spacetime about a stationary spherically symmetric body of total mass m. When we say stationary
we mean that it must be possible to introduce such a reference frame that the spatial location of
the body remains fixed forever. In such frame the components of metric tensor cannot depend on
time t. (Far away from the body this t must tick at the same rate as the proper time of observers
at rest relative to mass m. )
If the body is spherically symmetric then we expect the spacetime to be spherically symmetric as
well. Therefore, like in Euclidean space, we should be able to introduce spatial coordinates {r, , }
such that the line element depends on the angles and only via the combination

d2 + sin2 d2 .
Thus, we expect the metric form to have the following structure

ds2 = a(r, m)dt2 + b(r, m)dr2 + c(r, m)r2 (d2 + sin2 d2 ). (6.1)
There three unknown functions, they are a(r, m), b(r, m), and c(r, m), in this expression1 . It can be
reduced to two, if we redefine r via

(r0 )2 = c(r, m)r2 .


1 Notice that we introduced more than 4 additional constraints on the components of metric tensor assuming the

metric form (6.1) (see the discussion at the end of Sec.5.8)

69
70 CHAPTER 6. SCHWARZSCHILD SOLUTION

Then, (6.1) reads

ds2 = A(r, m)dt2 + B(r, m)dr2 + r2 (d2 + sin2 d2 ) (6.2)


where we have omitted 0 , or

A(r, m) 0 0 0
0 B(r, m) 0 0
g = .
0 0 r2 0
0 0 0 r2 sin2

Far away from this body we expect the curvature gradually reduce to zero. In other words, we
expect the spacetime to become flat at spatial infinity, that is

A, B 1 as r . (6.3)
In fact we can impose even more restrictive constraint on A(r, m). Indeed, given the results of
Sec.5.9, we may assume that far away from the body

A = 1 htt = 1 + 2 = 1 2m/r. (6.4)

(Here htt is the same as h00 in Sec.5.9)


This is how we think the metric tensor should look like in some suitable coordinates {t, r, , }.
It remains to be seen that Einsteins equations do indeed allow solutions of this form. The required
computations are carried out as follows: (i) compute g, , (ii) then g, , (iii) then R , (iv)
then R , (v) then G . To find the solution describing spacetime outside of the body we need to
substitute the result into the vacuum version of Einsteins equations:

G = 0.

This gives us a system of second order ordinary differential equations for A(r, m) and B(r, m) (note
that m is a parameter, not a variable.) which we need to solve subject to conditions at infinity. The
general solution of those equation is

A(r, m) = a(m) b(m)/r,

B(r, m) = c(m)/A(r, m).


Conditions (6.3,6.4) are satisfied by A(r, m) if

a(m) = 1, b(m) = 2m,

and the condition (6.3) is satisfied if


c(m) = 1.
Thus, the final result is

ds2 = (1 2m/r)dt2 + (1 2m/r)1 dr2 + r2 (d2 + sin2 d2 ) (6.5)

or
g = 0 if 6=
gtt = (1 2m/r), grr = (1 2m/r)1 , g = r2 , g = r2 sin2 .
This solution is known as the Schwarzschild solution and the coordinates {t, r, , } are called the
Schwarzschild coordinates. If the radius of the body is r then it holds only for r > r . However, the
Schwarzschild solution also describes the spacetime of a black hole in such case it applies for r > 0.
To be more precise, this solution applies only to non-rotating objects. Rotation inflicts additional
curvature on spacetime.
6.1. SCHWARZSCHILD SOLUTION 71

Let us analyse the nature of the Schwarzschild coordinates.

For any r > 0


~ ~ ~ ~
g = > 0, g = >0

and, thus, and are space-like coordinates.

For r > 2m
~ ~ ~ ~
gtt = < 0, grr = >0
t t r r
and, thus, t is a time-like coordinate and r is a space-like one as expected.

However, for r < 2m


~ ~ ~ ~
gtt = > 0, grr = <0
t t r r
and, thus, r is a time-like coordinate and t is a space-like one. Hence the lesson: Do not
assume that the coordinate denoted as t always refers to time measurements! Be prepared to
unexpected!

One can see that r = 0 is special. gtt and grr as r 0. In fact, the curvature scalar R
also tends to . At this point the curvature of spacetime becomes infinite. This is a real spacetime
singularity of the Schwarzschild solution the place where the approximation of General Relativity
breaks down.

6.1.2 Schwarzschild Solution in Kerr coordinates


r = 2m is also rather special as grr as r 2m. However, R remains finite and hence the
curvature of spacetime is finite. There is no spacetime singularity on this surface. In fact, on
this surface the system of Schwarzschild coordinates becomes singular. It is possible to introduce
other coordinate systems that are free from such singularity. One example is the system of Kerr
coordinates which is introduced as follows:

r, , and are the same as in Schwarzschild coordinates,

New t0 = t0 (t, r) coordinate is introduced via the following transformation, singular at r = 2m:

dt0 = dt (1 r/2m)1 dr, (6.6)

or
t0 t0
=1 = (1 r/2m)1
t r
Notice that Maxwells integrability condition

2 t0 2 t0
=
tr rt
is satisfied by the transformation (6.6)

Substitution of dt from (6.6) into (6.5) gives

ds2 = (1 2m/r)dt02 + (4m/r)dt0 dr + (1 + 2m/r)dr2 + r2 (d2 + sin2 d2 ). (6.7)

One can see that now all components of the metric tensor are finite at r = 2m and, thus, there is
no singularity there. Moreover, now the r-coordinate is always space-like. Notice, that eqs.(6.5) and
(6.7) describe the same spacetime (In what follows we will no longer use 0 to indicate Kerrs time.)
72 CHAPTER 6. SCHWARZSCHILD SOLUTION

6.1.3 Event horizon


Is it always possible to have a physical object, say a test particle, at rest relative to a black hole,
that is with fixed r, , coordinates? The spacetime interval along the world-line of any particle is
negative
ds2 = d 2 ,
where is the proper time of the particle. For a stationary particle
dr = d = d = 0
and, thus, along its world-line one has
ds2 = (1 2m/r)dt2
which is negative if r > 2m and positive if r < 2m. Thus, no stationary particle, as well as no
stationary physical observer, can exist at r < 2m!
If inside r = 2m particles must be moving then what kind of motion is it? It is easy to see that
all terms on the right side in eq.(6.7) are non-negative if r < 2m except the second one, which may
both positive and negative. Hence, if ds2 is negative then so must be this second term. This means
drdt < 0 and, hence, dr/dt < 0.
Thus the particle is forced to move inwards, toward the physical singularity at r = 0. The critical
radius rg = 2m is called the gravitational or Schwarzschild radius (in generic units rg = 2Gm/c2 )
and the surface r = rg is called the event horizon as nothing can escape from inside of this surface
into the outside space. Whatever event occurs inside the event horizon the outside observers are not
receiving any information about it.

Exercise
Determine the distance Lhs between the horizon and the singularity along the radial direction of
Kerr coordinates (t, , = const).

Solution: In Kerr coordinates


~ ~
= grr = 1 + 2m/r > 0
r r
and, thus, along the radial direction
ds2 = dl2 = grr dr2 > 0.
This is a space-like direction. Hence,
r=2m Z2m Z2mp

Z
Lhs = dl = grr dr = 1 + 2m/rdr
r=0 0 0
p
If we introduce new variable y = r/2m then
Z1 p
Lhs = 4m 1 + y 2 dy.
0

Given that Z p
1h p p i
1 + y 2 dy = y 1 + y 2 + ln(y + 1 + y 2 )
2
we finally obtain h i
Lhs = 2m 2 + ln(1 + 2) .
6.2. GRAVITATIONAL REDSHIFT 73

6.2 Gravitational redshift


Consider the Schwarzschild solution in Schwarzschild coordinates:

ds2 = (1 2m/r)dt2 + (1 2m/r)1 dr2 + r2 (d2 + sin2 d2 ). (6.8)


Consider an observer at rest (dr = d = d = 0) at infinity. If is the proper time of this observer
then

2
d = ds2 = dt2 . (6.9)
Thus, the coordinate t that selects the spacetime hypersurface t = const may be interpreted as the
time measured by an observer at rest at infinity by means of a standard clock.
Consider another observer at rest at 2m < r < . His/her proper time is

dr2 = ds2 = (1 2m/r)dt2 (6.10)

and, thus,

dr2 = (1 2m/r)d
2
(6.11)
Notice, that dr < d . This property is often described as slowing down of clocks ( or even
of time) in gravitational field. In fact, this is exactly what a distant observer watching a standard
clock of another observer, placed near a gravitating body, will see.
Consider two observers, A and B, resting at r = ra and r = rb respectively (both outside the
horizon). The interval of coordinate time t required for a light signal emitted by A to reach B
does not depend on the time of emission because the components of metric tensor in Schwarzschild
coordinates do not depend on t. To illustrate this point consider the case where both observers
are situated along the same radial direction (a = b , a = b . This simplifies the calculations.)
Due to the spherical symmetry of spacetime the light signal has to propagate along the same radial
direction and the spacetime interval along its world-line is given by

ds2 = gtt dt2 + grr dr2 = 0.

Therefore,
dt2 = (grr /gtt )dr2
and
r
Z b
p
t = (grr /gtt )dr . (6.12)

ra

Since g do not depend on t so does not t.


Suppose A emits two signals separated by the interval t of the coordinate time t. When B
receives these signals they are still separated by the same interval t. Indeed, if they are emitted
at t = 0 and t then they are received at t = t and t + t. For the same reason, if A emits a
periodic signal of period t of coordinate time t, B records the same period. However, the proper
time measured by standard clocks of the observers run at rates different from the rate of t. From
eq.6.10 one has
a2 = (1 2m/ra )t2 , b2 = (1 2m/rb )t2
and, thus,
 
1 2m/ra
a2 = b2 . (6.13)
1 2m/rb
If rb = then we have
74 CHAPTER 6. SCHWARZSCHILD SOLUTION

a2 = (1 2m/ra )
2
. (6.14)
Thus, if A emits a periodic signal with the period of its standard clock then B at r = will see
that this clock runs slower than his/her own standard clock. Notice, that eq.6.14 has exactly the
same form as eq.6.11.
On the other hand, could be just a period of a monochromatic electromagnetic wave emitted
by A as measured by his/her standard clock. Since the frequency of the wave = 1/ , we have
 
2 1 2m/rb
a = b2 . (6.15)
1 2m/ra
If rb > ra then b < a . Thus, the frequency of an electromagnetic wave is decreasing as the wave
propagates away from the source of gravity. This effect is called the gravitational redshift. (Optical
lines shift toward the red part of the spectrum).

6.3 Integrals of motion of free test particles in Schwarzschild


spacetime
By test particles we understand particles of such a small mass that their gravitational field is neg-
ligibly small compared to the field of other involved objects. Such particles can be used to test the
gravitational field created by those bodies without disturbing them. Hence the name test particles.
In Sec.5.3,5.7 we have learned that the equation of motion of a free particle is
Du
= 0, (6.16)
d
and, hence, its world-line is a geodesic of spacetime.
From Sec.4.4 ( just substitute with ) we know that the geodesic equations can be can be
written as the Euler-Lagrange equations

d L L
=0 (6.17)
d u x
with the Lagrangian

L(x , u ) = g (x )u u . (6.18)
These equations allows us to derive a number of very important results on the motion of test particles
in the Schwarzschild spacetime in a rather easy way.

Both in the Schwarzschild and Kerr coordinates

g
=0
t
and, thus,

L
= 0. (6.19)
t
From (6.17) and (6.19) one has
d L
= 0,
d ut
and, thus, dL/dut is an integral of motion, which means that it is constants along the world-line
of the particle. In fact,
L (g u u )
= = 2gt u = 2ut .
ut ut
6.3. INTEGRALS OF MOTION OF FREE TEST PARTICLES IN SCHWARZSCHILD SPACETIME75

Thus, we conclude that

ut = E = const. (6.20)
At infinity, where both in Schwarzschild and Kerr coordinates the metric attains its Minkowskian
form, one has

Ep
E = ut = gt u = gtt ut = ut = = (6.21)
mp
where Ep is the energy of the particle as measured by an observer at rest and mp is the rest
mass of the particle. For this reason E is called the specific energy at infinity.

Moreover, both in the Schwarzschild and Kerr coordinates


g
=0

and, thus,

L d L
=0 and = 0. (6.22)
d u
Since

L (g u u )

= = 2g u = 2u ,
u u
we conclude that

u = l = const (6.23)
is another integral of motion. It is called the specific angular momentum at infinity.

Since g depend on r and we conclude that ur and u are not integrals of motion!

However, a test particle with initial u = 0 placed in the equatorial plane, = /2 remains in
this plane forever. Since the direction of the polar axis is not restricted (spherical symmetry!)
this result simply tells us that the motion of free particles in Schwarzschild geometry is planar.
Let us derive this results. Consider the -component of (6.17):

d L L
= 0. (6.24)
d u

L (g u u ) g
= = u u =

(r2 sin2 )
= u u = 2 cos sin r2 (u )2 .

1 l
u = g u = l= 2 2
g r sin
Thus,
L 2 cos l2
= 2 3 . ()
r sin
Next,
L (g u u )
= = 2g u = 2g u = 2r2 u . ()
u u
76 CHAPTER 6. SCHWARZSCHILD SOLUTION

Substitution of (*) and (**) into (6.24) gives

cos l2
 
d d
r2 = 0. (6.25)
d d r2 sin3
It is easy to see that ( ) = /2 satisfies this equation. Moreover, this is the unique solution
satisfying the initial conditions

(0 ) = /2
d/d (0 ) = 0

(The theorem of uniqueness for second order ODEs.)

Exercise
A meteorite falls radially from rest at infinity into a Schwarzschild black hole. Show that in
Schwarzschild coordinates p
ur = 2m/r.

Solution
At infinity
ui = 0, =1 and, thus, E = 1.
Since the fall is radial, u = u = 0, the condition

g u u = 1

reads
gtt ut ut + grr ur ur = 1. (+)
t
Using E one can eliminate u from this equation. Indeed,
1 1
ut = g tt ut = E =
gtt gtt

Thus, eq.(+) reads


1 r r
gtt 2 + grr u u = 1.
gtt
Now we can find ur = ur (r):
1
(ur )2 = (1 + 1/gtt )/grr = (1 )(1 2m/r) =
1 2m/r

= ((1 2m/r) 1) = 2m/r.


6.4. ORBITS OF TEST PARTICLES IN THE SCHWARZSCHILD GEOMETRY 77

6.4 Orbits of test particles in the Schwarzschild geometry


Consider the Schwarzschild solution in the Schwarzschild coordinates:

ds2 = (1 rg /r)dt2 + (1 rg /r)1 dr2 + r2 (d2 + sin2 d2 ).

We already know that motion of test particles in the Schwarzschild spacetime is planar. We can
always choose the coordinates in such a way that the plane of motion becomes the equatorial plane,
= /2. Then u = 0 and the condition

g u u = 1

reads

gtt ut ut + grr ur ur + g u u = 1. (6.26)


Since
ut = g tt ut = E/gtt and u = g u = l/g ,
eq.(6.26) reads
E2 l2
+ + grr ur ur = 1.
gtt g
or
l2
E 2 + (1 + )gtt + grr gtt (ur )2 = 0. ()
g
For = /2 one has

g = r2 , grr = (1 rg /r)1 , gtt = (1 rg /r),

and eq.(*) reduces to


(ur )2 + (1 rg /r)(1 + l2 /r2 ) = E 2 .
Thus, we obtain
(ur )2 + l (r) = E 2 , (6.27)
where

l (r) = (1 rg /r)(1 + l2 /r2 ) (6.28)


From (6.27) one finds
dur 1 dl
= (6.29)
d 2 dr
This explains why is called the effective potential.
In the important case of a circular orbit
dur
ur = 0, and = 0,
d
and equations (6.27) and (6.29) reduce to

l (r) = E 2 , (6.30)
and
dl
=0 (6.31)
dr
respectively. These equations can be used to find the constants of motion, E and l, of circular orbits.
Before we proceed with the analysis of let us briefly review the Newtonian results.
78 CHAPTER 6. SCHWARZSCHILD SOLUTION

Newtonian theory
Similar analysis in Newtonian theory gives

(v r )2 + l (r) = E 2 , (6.32)
where
l (r) = 1 rg /r + l2 /r2 (6.33)
Let us figure out how the motion of a particle with the specific angular momentum l looks like
in the plane E 2 against r. Since E is an integral of motion the particles move parallel to the
r-axis. From (6.32) it follows that
E 2 l (r) 0
and, thus, their motion is confined within the region above the curve E 2 = l (r). Everywhere
on this curve v r = 0 but only the extremum corresponds to a circular orbit (see condition
(6.31). All other points of this curve are turning points.

From this figure it follows that


No particle with l 6= 0 can ever reach r = 0.
Particles with E 2 > 1 will always escape to infinity, even if their initial vr < 0.
Particles with E 2 < 1 will move between r and r+ .
Einsteins theory
Differentiating (6.28) one obtains
l2
 
dl rg 2 2
= 4 r 2 r + 3l
dr r rg
Thus, the extrema of l are the solutions of
 2
l
r2 2 r + 3l2 = 0, (6.34)
rg
The solutions to this quadratic equation are

l2
q
r = l (l/rg )2 3. (6.35)
rg
Thus,
6.4. ORBITS OF TEST PARTICLES IN THE SCHWARZSCHILD GEOMETRY 79

If l2 > 3rg2 then there are two circular orbits with radii r+ and r ,
If l2 = 3rg2 then there is only one circular orbit with the radius rms = 3rg ,
If l2 < 3rg2 then there are no circular orbits.

Once again we can understand the properties of orbits by using the E 2 -r plane. The figure
below shows the curves E 2 = l (r) for various values of l. Notice that now l 0 as r 0.

From this figure it follows that


Now there exist trajectories leading directly to singularity. Particles may be swallowed
by a black hole.
For l2 > 3rg2 there exist oscillating orbits, for l2 < 3rg2 such orbits do not exist.
The circular orbit with r = r+ is stable, whereas the one with r = r is unstable.
There are no stable orbits with r < rms = 3rg . The orbit with r = rms is called the last
stable orbit.

Exercise 1:
Determine the integrals of motion of circular orbits.

Solution:
From eq.(6.34) one has

mr2
l2 = . (6.36)
r 3m
From eqs.(6.30,6.28,6.36) one has

E 2 = l (r) = (1 2m/r)(1 + l2 /r2 ) =

(r 2m)2
= (1 2m/r)(1 + m/(r 3m)) = .
r(r 3m)
Thus,

(r 2m)2
E2 = . (6.37)
r(r 3m)
80 CHAPTER 6. SCHWARZSCHILD SOLUTION

Exercise 2:
A spaceship is orbiting a black hole of mass m. Given that its orbit is circular one with radius r
determine the orbital period as measured by
(i) a passenger of the spaceship, T(i) ,
(ii) a stationary observer far away from the hole (at infinity), T(ii) .

Solution:
(i) The period T(i) is measured by a standard clock carried with the ship. Its time is the proper
time of the ship, . Since
d
u =
d
one has
T(i) = 2/u .
But 1/2
mr2

1
u =g u = l/g = 2 =
r r 3m
 1/2
1 m
= .
r r 3m
Hence,
 1/2
r 3m
T(i) = 2r .
m

(ii) The period T(ii) is measured by a standard clock at rest at infinity. It runs with the same
rate as t (see Sec.6.2). Hence,
dt
T(ii) = T(i) = ut T(i) .
d
But ! r
t tt 1 r 2m r
u = g ut = E/gtt = p = .
1 2m/r r(r 3m) r 3m
Thus, r
r
r r 3m p
T(ii) = 2r = 2r r/m.
r 3m m

6.5 Perihelion shift of planets


The biggest success of Newtonian theory of gravity is the great accuracy with which it describes the
motion of planets around Sun. For example, it predicts that the planetary orbits are not exactly
elliptical. The closest point of a planetary orbit, called perihelion, gradually moves (precesses) in
the orbital plane due to the gravitational interaction with other planets. This is indeed what is
observed. However, by the beginning of last century a small annoying disagreement between the
theory and measurements of the precession of the closest planet to Sun, Mercury, was becoming
apparent. General relativity explains this result.
Let us find the equation for the trajectory of test particle in Schwarzschild geometry. Select
Schwarzschild coordinates in such a way that the orbital plane has = /2. From eq.6.27 we have

(ur )2 = E 2 (1 rg /r)(1 + l2 /r2 ).

Next,
l l
u = g u = = .
g r2
6.5. PERIHELION SHIFT OF PLANETS 81

Combining these two results we find


 2  r 2
dr u E 2 (1 rg /r)(1 + l2 /r2 )
= = . (6.38)
d u l2 /r4

Next we introduce more convenient variable


rg
w=
r
(recall that rg = 2m in geometric units and rg = 2Gm/c2 in generic units). Then eq.6.38 reduces to

(w0 )2 = E 2 a (1 w)(a + w2 ), (6.39)

where w0 = dw/d and


 r 2
g
a= .
l
Next differentiate eq.6.39 to obtain
3 a
w00 + w w2 = . (6.40)
2 2
This is the equation of particle trajectory. Circular orbits are characterized by vanishing w00 . Thus,
for circular orbits
3 2 a
w w + = 0.
2 2
The typical distance of Mercury from Sun is r ' 5.5 107 km, whereas Suns gravitational radius is
only rg ' 2.94km. Thus, w is very small, ' 5 108 . Then, we can ignore the quadratic term in
the above equation and write
a
w= .
2
This tells us that a is also very small. Let us now consider slightly non-circular orbits
a
w= + s,
2
where s  a  1. Substitute this into eq.6.40 and retain only terms zero and first order in s:

3a 3a2
s00 + (1 )s = . (6.41)
2 8
The general solution of this linear ODE is easy to find,

3a2
s= + A cos( b + B),
9
where A and B are arbitrary constants and
3a
b=1 .
2
Thus, we obtain
a 3a2
w=( + ) + A cos( b + B). (6.42)
2 9
The Newtonian theory gives somewhat different solution to this problem,
a
w= + A cos( + B). (6.43)
2
This equation describes perfect ellipse. The fact that the orbit is a closed curve comes from the fact
that the phase, = + B, of cos-function in eq.6.43 changes exactly by 2 when changes by 2.
82 CHAPTER 6. SCHWARZSCHILD SOLUTION

perihelion
p

Figure 6.1:

The appearance of term 3a2 /9 in the relativistic result tells us that the radius of circular orbit for
a particle with given angular momentum differs from value given by Newtonian theory. Moreover,
since b 6= 1 non-circular orbits are no longer closed curves. They can be described as elliptical orbits
with precessing perihelion (see fig.6.1).
Indeed, the planet reaches its perihelion each time when the phase, = b + B, of cos-function
in eq.6.42 increases by 2. The corresponding increase in
3 3
= 2/b = 2(1 a)1/2 ' 2(1 + a)
2 4
differs from 2 by the value
3
p = a. (6.44)
2
This is the angular shift of the perihelion per one orbital turn. Substituting the data for Mercury
one finds
p = 5 107 radian per turn = 43 arcsec per century.
This is exactly the shift that could not be accounted for in Newtonian theory. This was the first
positive experimental test of General Relativity.

6.6 Bending of light


Photons are different from massive particles as they move with the speed of light. Since no clock
can move with the speed of light proper time can no longer be used as a parameter along their
6.6. BENDING OF LIGHT 83

world lines. This leads to slight modification in the analysis of their motion. When photons are
non-interacting with other particles this motion is still geodesic and it is still governed by the Euler-
Lagrange equations
d L L
=0 (6.45)
d u x
with the Lagrangian
L(x , u ) = g (x )u u .
However, now is just some normal parameter, which we do not have to specify in advance, and
dx
u =
d
is the corresponding tangent vector to photons world-line (null vector). Repeating exactly the same
calculations as those in Sec.6.3 we derive the same results as for massive particles

ut = E = const, (6.46)

u = l = const, (6.47)
and
cos l2
 
d d
r2 = 0. (6.48)
d d r2 sin3
Just like in the case of massive particles the last equation tells us that the motion of photon is still
planar and without any loss of generality we may assume that it occurs in the equatorial plane of
Schwarzschild coordinates:
= /2, u = 0.
Now ~u is null,
g u u = 0,
and in Schwarzschild coordinates this reads

gtt (ut )2 + g (u )2 + grr (ur )2 = 0

or
gtt (ut /u )2 + g + grr (ur /u )2 = 0. (6.49)
Since
ut = g tt ut = E/gtt ,
u = g u = l/g ,
ur /u = dr/d
eq.6.49 reduces to
2  2  2
g E dr
+ g + grr = 0.
gtt l d
Substituting expressions for the components of Schwarzschild metric, denoting l/E as b, and intro-
ducing new variable
w = rg /r,
which is more convenient in the following calculations, we obtain
 2  r 2
dw g
+ w2 (1 w) = 0. (6.50)
d b
Finally, we differentiate this equation with respect to to get

d2 w 3
2
+ w w2 = 0. (6.51)
d 2
84 CHAPTER 6. SCHWARZSCHILD SOLUTION

This second order ODE determines the space trajectory of a photon (w = w() or r = r()), given
its initial position, w(0 ), and initial direction of motion, dw/d(0 ). (Given w(0 ) and dw/d(0 )
one can find the value of parameter b for this photon from eq.6.50.)
Next we find the perturbative solution for the trajectory of photon in the warped space-time of
the Sun and show that this trajectory is bend. This bending is just large enough to be measured
using the technology available at the beginning of the last century.
Let the initial conditions be
dw
w(0) = rg /r(0) = a, (0) = 0 (6.52)
d

(the latter means that at = 0 the photon trajectory is perpendicular to the radial direction). The
closest distance a photon can get to the Sun is Suns radius, r ' 7 1010 cm. Suns gravitational
radius is much smaller, rg ' 3 105 cm. Thus, the highest possible value of a is only about 5 106 .
This means that we have a small parameter in this problem and can use perturbative approach. We
shell seek solution in the form of asymptotic expansion

w() = aw1 () + a2 w2 () + a3 w3 () + ... . (6.53)

In fact, we will only find the first two terms in this expansion and ignore the rest as a small
contribution. Substitute this expression into eq.6.51 and ignore all terms that involve a to the power
of 3 and higher. The result is

3
aw100 + a2 w200 + aw1 + a2 w2 a2 w12 = 0. (6.54)
2
Next we collect terms of the same order in a
3
a(w100 + aw1 ) + a2 (w200 + w2 w12 ) = 0. (6.55)
2
One can see now that w1 can be found via integrating

w100 + aw1 = 0 (6.56)

and then w2 can be found via integrating


3
w200 + w2 w12 = 0. (6.57)
2
The initial conditions (6.52) and the expansion (6.53) imply the following initial conditions for w1
and w2 :
w1 (0) = 1, w10 (0) = 0, (6.58)
and
w2 (0) = 0, w10 (0) = 0. (6.59)
We are dealing with simple second order linear ODEs. The solution of (6.56,6.58) is

w1 () = cos ,

whereas the solution of (6.57,6.59) is

1 1
w2 () = 1 cos cos2 .
2 2
Thus, the second order perturbative solution is
1 1
w() = a cos + a2 (1 cos cos2 ). (6.60)
2 2
6.6. BENDING OF LIGHT 85

y r

Figure 6.2:

If we kept only the terms of first order in a then we would obtain

r cos = rg /a = const.

This describes trajectory which runs parallel to the y-axis in the equatorial plane (see figure 6.2 )
and hence shows no bending.
In order to find the magnitude of total bending described by eq.(6.60) consider the limit r =
or w = 0. Then from eq.(6.60) we find

1 cos2
cos = a . (6.61)
1 a/2

Since a  1, this equation tells us that with great accuracy we can put

cos = a.

The two solutions to this equation can be written as = (/2 + ) where > 0

sin = a.

Because a  1 this implies that = a with great accuracy. Then the total bending angle is given
by
= 2 = 2a. (6.62)
For a photon that passes just above Suns surface r(0) = r = 6.97 1010 cm. This gives us the
highest possible value of a = 4.2 106 and

= 1.75arcsec.

In 1919 British astrophysicist A.Eddington confirmed this prediction via observations of background
stars during full solar eclipse. The result propelled General Relativity to worldwide recognition.
86 CHAPTER 6. SCHWARZSCHILD SOLUTION
Chapter 7

Appendix

7.1 Geometric units


It is well known that physics needs 3 basic units, for length, time, and mass, and that all other units
could be derived from these three. Any particular set of units gives certain values for the universal
physical constants, like c, G, and h. However, the universal physical constants allow to reduce the
number of independent physical units. For example, instead of usual time t one can introduce new
time variable t0 = ct, the geometric time. Obviously, this new time variable will have the dimension
of length,
[t0 ] = L.
Similarly, instead of usual mass, m, one can introduce the geometric mass, m0 = Gm/c2 . It also has
the dimension of length,
[m0 ] = L.
Once, we have opted to use geometric time and mass, all other physical quantities get dimension
which is a power of L (L0 , L, L2 etc.). For example, the dimension of force, which is defined via
dx
f 0 = m0 ,
dt02
is
L
[f 0 ] = L = L0 .
L2
Thus, force becomes dimensionless.
Summarizing, the transition to geometric units is described by

t0 c2 0
t , m m. (7.1)
c G
and the inverse transition to generic units is described by
G
t0 ct, m0 m. (7.2)
c2
These also allow to relate the magnitudes of various physical variables in generic and geometric
units. Take force for example. According to Newtons definition

d2 x c2 d2 x c4 d2 x c4
f =m 2
= m0 02 2 = m0 02 = f 0 .
dt G dt /c G dt G

What can we say about the dimensions of G0 and c0 ? The dimension of speed in geometric units
is
L
[v 0 ] = = L0 .
L

87
88 CHAPTER 7. APPENDIX

Thus, speed becomes dimensionless and so is the speed of light. The dimension of G0 becomes
apparent when we inspect Newtons law of gravity

m01 m02
f 0 = G0 .
r2
Indeed, as we have seen [f 0 ] = L0 and

m01 m02
 
LL
= 2 = L0 .
r2 L

Thus,
[G0 ] = L0 .
The gravitational constant is dimensionless as well.
Can we say anything about the values of G and c in geometric units? Yes,

G0 = 1 and c0 = 1. (7.3)

Indeed, in a locally inertial frame in generic units we have

ds2 = c2 dt2 + dl2 .

In generic units this reads


dt02
ds2 = c2 + dl2 = dt02 + dl2 .
c2
This tells us that c0 = 1. Next, consider the Newton gravity law
m1 m2
f =G .
r2
Converting to geometric units we obtain

c4 0 m0 (c2 /G)m02 (c2 /G) c4 m01 m02


f =G 1 = .
G r2 G r2
or
m01 m02
f0 = .
r2
This tells us that G0 = 1.

You might also like